AI
Dr. Jon KrohnIn Case You Missed It in April 2026 (Ep. 988 with Jon Krohn)
synopsis
The host's April 2026 roundup runs from theory to practice to learning. Agents now need memory modelled on human memory. They are taking over the junior work through which people used to enter data careers. What remains valuable is systems thinking, domain depth, and AI literacy taught from childhood.
Agent memory (episode 985). Richmond Alake of Oracle sets out four memory types. Episodic memory is time-stamped events such as conversation turns; the host gives the example of a travel agent recalling a Monday client visit. Procedural memory is routines like skills.md files, which Alake calls SOPs for agents, and the host compares it to knowing how to scramble eggs without remembering learning it. Semantic memory is facts and institutional knowledge, and working memory is the context window. Alake says Oracle is testing skills stored in database tables with vector representations so agents retrieve only the ones needed.
Data engineering agents (episode 981). Matt Glickman of Genesis Computing says the company augments people rather than platforms. Current staff keep their jobs and gain tenfold capacity, but junior hiring is what AI will wipe out. The host suggests juniors raised on vibe coding might become skilled agent "conductors". Glickman hopes so, but says most schools still discourage AI and that telling AI-written work apart is "a farce".
Infra and transferable skills (episode 987). Linda Haviv says theoretically no code is now needed to build AI, yet production scale brings back systems thinking. Non-technical builders describe infrastructure problems like memory and lag in their own words. She argues engineers know where things fail, domain experts know their niche, and people should bring their whole background, as she did from TV news to building a newsroom CMS.
AI in a K-8 school (episode 983). Tracy Walker Griffith describes fifth graders comparing their own test answers with AI's in Magic School and tracking their mistakes so teachers spot patterns. Her line is "AI raised the floor. Teachers are raising the ceiling," with AI literacy as the foundation.
- Agent memory maps onto four kinds of animal memory. Richmond Alake, Oracle's director of AI developer experience (episode 985), splits agent memory into four types. Episodic memory is time-stamped, like conversation turns with role, content and timestamp. Semantic memory is world and institutional knowledge. Procedural memory is routines such as skills.md files, which he calls "SOPs for agents". Working memory is the LLM's context window. He and the host, who studied neuroscience, ground this in biology's role in inspiring CNNs through Hubel and Wiesel's experiments on the cat visual cortex.
- Storing skills in a database lets agents retrieve only what they need. Alake says Oracle is experimenting with putting skills in database tables, including vector representations, rather than files. Skills can then be exposed progressively instead of loading 100 skills into context, which he says allows scaling.
- Data engineering agents will curb junior hiring, not cut existing staff. Matt Glickman, CEO of Genesis Computing (episode 981), says no one is losing their job: existing data engineers were the limiting factor and now get "10x" more done. Junior hiring is what AI will "wipe out", and it is already starting. A few hires will be "conductors" running many agents in parallel; Glickman says a co-founder runs six while Glickman manages two and a half. Most schools still discourage AI use, which Glickman calls a "double whammy" for graduates.
- Coding is being abstracted, but AI infrastructure and systems thinking are surfacing. Linda Haviv (episode 987) notes non-technical people now talk about memory and context lag, which are infrastructure problems in other words. Engineers' edge, she argues, is knowing where systems fail, since even model training is an orchestration problem. People who know a domain should bring that context, as she did moving from media to building a news CMS.
- AI raised the floor in one school; teachers are raising the ceiling. Tracy Walker Griffith, principal of a K-8 innovation school in Boston (episode 983), describes fifth graders preparing for the state test. They write an answer, have AI in Magic School write one, explain which is better, and log wrong multiple-choice answers in a tracker the teacher checks for patterns. She says the school stays "platform-informed, not platform-loyal", also using PlayLab, which runs on Claude. AI literacy for both teachers and students comes first.
Professor G Markets$The Case for AI Optimism — with Reid Hoffman
- This post includes paid content that is not available publicly. See the original source.
Professor G Markets$Trump Wins the Battle for Attention, Loses the War in Iran
- This post includes paid content that is not available publicly. See the original source.
Professor G Markets$The Iran War Has No Exit — ft. Ian Bremmer
- This post includes paid content that is not available publicly. See the original source.
Professor G Markets$Ray Dalio: The World Order Has Unraveled
- This post includes paid content that is not available publicly. See the original source.
Dr. Jon KrohnWhy Your Side Hustle Is the Best Career Insurance You'll Ever Have (with Linda Haviv)
- Work on the side is insurance against being a number. Haviv argues that however good the people are, to the company as a whole you are "a human resource", so she never kept all her eggs in one basket; she chose software partly because it is a skill no employer can take away.
- Her personal brand fed her full-time job. Haviv says her job at Amazon came about because the company found her talking about cloud computing on TikTok, and that the side work made her better in the day job rather than distracting from it.
- Independent work lets her be more intentional than inside a company. She says committing to a single project or campaign without internal red tape is more efficient for her, and that she can keep advocating for tools she likes (Ray, AWS) from outside; the host lists her independent brand work as including Anthropic, Dell, Amazon, GitHub, Microsoft and Nvidia.
- Work is getting leaner, toward tiny billion-dollar companies. Haviv likens the present to the arrival of the internet and guesses that maybe 50% of jobs will be at small companies, with two- to three-person and perhaps one-person billion-dollar companies; she says plainly this is her own impression, "not some scientifically backed metric".
Professor G Markets$Trump’s Base Feels Betrayed by Another Forever War, as America Spirals Into Conspiracies — with Astead Herndon
- This post includes paid content that is not available publicly. See the original source.
Professor G Markets$Data Center Debate: Are Energy Bills About To Explode?
- This post includes paid content that is not available publicly. See the original source.
Dr. Jon KrohnWhy AI Infrastructure Is Everyone's Problem Now (with Linda Haviv)
- Non-technical builders now hit AI-infrastructure problems in other words. Haviv says that once people move past toy examples to production and scale, they run into systems issues; non-technical people now talk about memory, context that "is not working" and lag, which are the same problems as KV cache and prefill-decode disaggregation without the vocabulary.
- Engineers' edge is knowing where things fail. She argues that agentic workflows sit on a distributed-systems layer, and that even model training and mixture-of-experts models are orchestration problems, so someone from DevOps or software brings real strength even without AI/ML expertise. The host's premise is that within "a couple months" it has become theoretically possible to do the whole AI pipeline in natural-language prompts.
- Domain knowledge transfers, so start where you already understand users. Haviv, who taught herself to code in 2015 and took her first JavaScript job in 2016, says her media background let her build a content management system for journalists because she had written the articles herself; the host draws the lesson that a newcomer may do better moving into development at their current employer than applying to Meta.
- Building and media are democratised, so curate what you build. She says creators no longer need a $50,000 studio or a label, and that she saw in TV news that specialists such as lawyers and doctors who came on as experts kept control of their time. The host adds that hands-on experience in a field gives more confidence in AI advice and warmer contacts than a cold email; Haviv notes junior developers struggling to find jobs out of college are also turning entrepreneurial.
Professor G Markets$Trump Blames Democrats, Demands His Ballroom, and Attacks Jimmy Kimmel Again — with Sen. Rand Paul
- This post includes paid content that is not available publicly. See the original source.
Professor G Markets$REPLAY: Building AI Supercompanies with Greg Shove and Scott Galloway
- This post includes paid content that is not available publicly. See the original source.
Professor G Markets$The Biggest IPO In History Isn’t What You Think It Is
- This post includes paid content that is not available publicly. See the original source.
Dr. Jon KrohnEx-Anyscale: 1-Person Billion-Dollar Companies Are Coming (Ep. 987 with Linda Haviv)
synopsis
Linda Haviv, who at the time of recording had just left her role as staff developer advocate at Anyscale to work independently on AI-infrastructure education, content and developer-relations consulting, ends up arguing that this is the best moment to build on one's own: coding is being abstracted away, but the systems problems underneath it, the AI-infrastructure layer, are surfacing for everyone, and people who bring transferable skills, open-source habits and a community will do best. The second half of the conversation turns technical, on Ray, what "AI infra" means and open-source governance.
A non-linear path. Haviv majored in philosophy at CUNY Baruch because it "does not have an answer", worked in TV news before law school, and taught herself to code by building book sites for anchors. She found developer relations through New York meetups, took her first JavaScript job in 2016, and has made technical content on Instagram since then. She says the profession is now going through an identity crisis: she misses writing code, though watching agents work has its own satisfaction, and systems thinking, once a principal-level skill, is now needed by more people.
Infrastructure surfaces for everyone. The host's premise is that within "a couple months" it has become theoretically possible to train, fine-tune, deploy and front an AI application through natural-language prompts alone. Haviv accepts the speed but says that beyond toy examples people hit systems problems. What stopped her that week was a non-technical person talking about memory: builders complain that context "is not working" or lags, which is the KV-cache and prefill-decode-disaggregation problem in different words. Engineers keep an edge because they know where things fail; agentic workflows sit on a distributed-systems layer, and even model training is an orchestration problem. Domain knowledge transfers as well: she could build a content management system for journalists because she had written articles. The host draws the lesson that a newcomer may do better moving into development at a current employer than applying to Meta first.
Community and building in public. Both prefer in-person learning; Haviv attended Flatiron School at the height of the bootcamps and expects in-person events to grow in popularity as more work goes digital. She says you learn by reading others' code, and that explaining something when you are one step ahead is often the clearest version for the next learner. Her advice is to start small, with a screenshot and a caption.
Why go independent now. Haviv calls side work insurance: to a company you are "a human resource", and her Amazon job came because the company found her cloud-computing videos on TikTok. She likens the moment to the arrival of the internet and guesses, explicitly without data, that maybe 50% of jobs will be at small companies, with two- to three-person and perhaps one-person billion-dollar companies. Outside a company she finds project work more efficient and less tied up in red tape. She plans content, developer-relations consulting focused on digital communities and technical founders' personal brands, and building her own systems on Claude.
Ray and AI infrastructure. Haviv describes Ray as a Python-native open-source distributed computing framework, built in Ion Stoica's Berkeley lab by Robert Nishihara and colleagues because ML researchers were spending more time on distributed systems than on reinforcement-learning research. It now spans Ray Data, Ray Train and Ray Serve, integrates natively with vLLM and Hugging Face, and suits multimodal workloads that do not fit on one GPU better than Spark does. She defines AI infrastructure loosely, as infrastructure specific to compute-heavy AI/ML workloads, overlapping with traditional infrastructure; inference is where users first meet it. Neoclouds such as Nebius, CoreWeave and Crusoe are specialists next to hyperscalers that "do everything". Her popcorn analogy for batch inference has single kernels as latency-optimised and the always-on microwave as throughput-optimised, with CPU preprocessing needed to keep the GPUs fed.
Open source and governance. She says it is crucial that Ray, vLLM, Kubernetes and MCP sit under Linux Foundation governance rather than with one company, and expects open models such as the newly released Gemma 4 and DeepSeek to catch up. She runs Gemma 4 on her phone for offline use and says she has moved most of her assistant work from OpenClaw to Claude skills and plugins.
- Leaner work makes this the time to go independent. Haviv, who had just left her staff developer advocate role at Anyscale at the time of recording, guesses that maybe 50% of jobs will be at small companies and that two- to three-person, perhaps one-person, billion-dollar companies are coming; she flags this as her own impression, not a measured figure. She treats side work as insurance against being "a human resource", noting her Amazon job came from the company finding her cloud-computing videos on TikTok.
- AI-infrastructure problems are now everyone's problems. Haviv says non-technical builders now complain about memory, context and lag, which are infrastructure problems in different words, and that past toy examples the work becomes systems thinking. She argues engineers keep an edge because they know where things fail, since even model training and mixture-of-experts are orchestration problems; the host frames the premise that the whole AI pipeline has become doable in natural language within "a couple months".
- A Python-native framework spares ML engineers distributed-systems expertise. Haviv describes Ray as a Python-native open-source distributed computing framework, created at Ion Stoica's Berkeley lab by Robert Nishihara and others for reinforcement learning, now spanning Ray Data, Ray Train and Ray Serve with native vLLM and Hugging Face integrations. She says it suits multimodal, GPU-bound AI workloads better than Spark because those workloads do not fit on one GPU.
- AI infrastructure is infrastructure shaped by compute-heavy AI workloads. She defines it as infrastructure specific to AI/ML workloads, which are compute-heavy where older workloads leaned on data input and output, and says inference is where users first hit the difference. Her popcorn analogy for batch inference: popping one kernel is latency-optimised, a microwave full is throughput-optimised but costs money while on, so CPU preprocessing must keep the GPUs fed. She casts neoclouds (Nebius, CoreWeave, Crusoe) as specialists beside hyperscalers that "do everything".
- Open governance of the AI stack matters more than open weights. Haviv says it is crucial that Ray, vLLM, Kubernetes and MCP sit under the Linux Foundation rather than being owned by one company, and expects open models such as the newly released Gemma 4 and DeepSeek to catch up with proprietary ones. She runs Gemma 4 on her phone for offline use on flights.
Professor G Markets$The U.S. vs China AI Battle Is Getting Ugly
- This post includes paid content that is not available publicly. See the original source.
Professor G Markets$The Aftermath of Trump's Third Assassination Attempt
- This post includes paid content that is not available publicly. See the original source.
Professor G Markets$Why International Stocks Are Beating the S&P, and How Scott Invests his Money
- This post includes paid content that is not available publicly. See the original source.
Professor G Markets$The Longevity Economy Is Built for the Rich
- This post includes paid content that is not available publicly. See the original source.
Dr. Jon KrohnThe Most Obvious Agent Component Everyone Ignored (with Richmond Alake)
- Memory was the obvious agent component everyone left out. Alake says that two years ago definitions of agents listed tools, reasoning and planning but never memory, because the labs were focused on top-end benchmarks and the coding-agent use case. He calls memory "the final battleground", and stresses that the underlying ideas, such as continual learning, are old rather than new.
- Developers want memory they never have to think about. Alake says developers he spoke with are waiting for a reliable system that handles both retrieval and forgetting, as human memory does without effort. He expects it from research labs and database companies, but says this is one field where a single developer could build something that competes with a lab's solution. The host adds that grouping memory in as just another tool undersells it.
- The agent stack will consolidate because tool selection isn't where value lies. Alake compares today's fragmentation to the old Angular-versus-React battles of web development, and predicts a few staple players, as with the small number of key model vendors. He says his employer Oracle intends to put heavy resources into agent memory so that developers "don't need to shop around".
Professor G Markets$Two Months In: Why Markets Stopped Caring About Iran
- This post includes paid content that is not available publicly. See the original source.
Dr. Jon KrohnThe Hidden Cost of Using Too Many Databases for AI Agents (with Richmond Alake)
- The field now calls the assembled agent stack an "agent harness". Alake lists the layers that matter as orchestration, connectivity, security, reasoning, a memory core where data is stored, and memory managers, and says the stack is expanding and consolidating at once. He cites Claude Code and Codex as harnesses built around specific models, and says he published a piece on effective harnesses in May of the previous year.
- Build memory-first: assume everything must be recalled or forgotten. Alake argues that starting from the premise that every piece of incoming information must later be recalled or forgotten changes how a team models its data and chooses tools.
- Multiple databases are an anti-pattern because they add cognitive load. He says the developer's job is to reduce the LLM's cognitive load so it can focus on reasoning, and that gluing several databases together adds load for both the model and the engineers. He argues speed of experimentation is the moat in AI, so time spent choosing databases and query languages is wasted.
- Own the domain workflow and buy the general-purpose plumbing. Alake says he could never build a healthcare or legal agent better than an engineer inside that industry, so engineers should study the workflow they are automating in detail. For storage he pitches his employer's Oracle AI Database, which he says handles vector, graph, relational, lexical and spatial data and can run embedding models inside the database.
Dr. Jon KrohnWhy RAG Alone Isn't Enough for AI Agents (with Richmond Alake)
- RAG covers retrieval, but memory also needs updating and forgetting. Alake defines RAG as retrieving information and passing it with the user's prompt to ground the model in domain data, and says agent memory must also update, consolidate, resolve conflicts between and forget memories. He treats RAG as one tool in the agent-memory toolbox.
- Conflicting source documents break naive RAG. The host's example comes from his previous HR-software startup: HR manuals written by different managers in 2023 and 2024 can give opposite instructions for the same incident, so retrieval can tell an employee one thing on day one and the opposite on day two, each time from "correct" information.
- Resolve conflicts offline, not in working memory. Alake says conflict resolution is computationally expensive and should not sit in working memory, because users should not wait while the system works out which memory to use; the result should be stored permanently.
- Old database techniques still matter for memory latency. He cites quantization, lowering the precision of vector embeddings to speed retrieval and cut storage, as a decades-old technique that someone vibe coding an AI application today would not know. He argues newcomers need educating while long-time database people are brought along too.
Professor G Markets$A War That Trump Won’t End — and an Economy Starting to Feel It
- This post includes paid content that is not available publicly. See the original source.
Professor G Markets$The Case for Making Up with China, and Which Car Company Is Winning the Energy Crisis?
- This post includes paid content that is not available publicly. See the original source.
Dr. Jon KrohnBuilding Hardware is Hard but AI Agents Help (Ep. 986 with Kishore Subramanian)
- Product lifecycle management carries hardware from concept to customer. Subramanian, Propel Software's CTO and formerly a senior engineer on Google Assistant, describes its cloud product lifecycle management (PLM) and quality management system (QMS) as the system of record for a physical product from concept to customer and beyond (bills of materials, manufacturing information, compliance, repairs), calling Propel "Atlassian for physical products". Its main verticals are high-tech, med-tech and consumer goods. The host says up front that the episode is a partnership with Salesforce.
- Agents reviewing engineering change orders shift hardware quality left. Subramanian calls an engineering change order the hardware equivalent of a pull request, and says Propel One's change-review and quality agents flag downstream risk and compliance gaps the way code-review agents do for software. The host adds that hardware cannot simply be patched, citing car recalls of millions of vehicles; Subramanian adds supply-chain shocks, such as a part becoming unavailable or pricier for geopolitical reasons, that threaten manufacturing deadlines.
- Building on a vendor's agent platform let Propel ship fast. Propel is an ISV on the Salesforce platform, and Subramanian credits its extensibility of data model, process and user experience for serving different verticals, including med-tech's FDA and audit demands. Propel One, begun around January 2025 and released to existing customers within six months, by mid-2025, uses Agentforce for security and governance over customers' core product IP, plus Data 360, Agent Builder, Agent Script and the Atlas reasoning engine.
- Open agent tooling is not yet enterprise-mature, he says. When the host raises known security issues with MCP and OpenClaw, Subramanian says Propel does not rule them out but will adopt them once they move past the prototype stage. He adds that much tuning happens at implementation because every customer differs, which is why forward-deployed engineers are growing in popularity.
- To get agents into production, start from how a human does the task. Subramanian's advice is to break the human process into tasks, decide which AI can automate, check that the data exists, build the underlying tools as code, and design security in from the start rather than as an afterthought.
Professor G Markets$The AI Job Crisis Andrew Yang Saw Coming
- This post includes paid content that is not available publicly. See the original source.
Professor G Markets$Trump Crashes Out as “Madman” Iran Strategy Blows Up
- This post includes paid content that is not available publicly. See the original source.
Professor G Markets$America Has a Moral Problem, Not a Political One — with David Brooks
- This post includes paid content that is not available publicly. See the original source.
Professor G Markets$The Most Dangerous AI Model Just Leaked
- This post includes paid content that is not available publicly. See the original source.
Dr. Jon KrohnThe Four Types of Memory Every AI Agent Needs (with Richmond Alake)
- Agent memory borrows its categories from biology. Alake splits it into episodic, procedural, semantic and working memory, arguing technology has always taken inspiration from nature. The host, who has a neuroscience background, retells the Hubel and Wiesel cat visual-cortex experiments, in which neurons fired only for straight-line edges at a specific orientation and were later found to combine into edges, corners and more abstract layers, as convolutional networks do.
- Episodic memory is timestamped events, such as conversation turns. Alake says each message stored with its role (user or assistant), content and timestamp is episodic, and that the timestamp is what makes it recallable. The host's analogy is a travel agent remembering that a client came in at 10 a.m. on Monday.
- Procedural memory is skills, which can be retrieved from a database. Alake describes skills.md files as standard operating procedures for agents, the analogue of routines the cerebellum stores. He says Oracle is experimenting with storing skills in database tables with vector representations so an agent retrieves only the skills it needs instead of loading 100 skill files into context, which lets it scale.
- Semantic memory is facts; working memory is the context window. He says semantic memory is world or institutional knowledge, such as an enterprise's abbreviations and how things are done, held as facts tied to no episode or procedure. Working memory, a subset of short-term memory used in real time, maps to the LLM's context window.
Professor G Markets$Dems Beat Back Trump’s Redistricting Agenda to Steal the Election — with Carlos Espina
- This post includes paid content that is not available publicly. See the original source.
Professor G Markets$Tim Cook Sat Out the AI Race — Will the New CEO Pay the Price?
- This post includes paid content that is not available publicly. See the original source.
Dr. Jon KrohnWhy "Memory Engineer" Is the Next Big Role in AI (with Richmond Alake)
- "Memory engineer" bridges database veterans and today's AI engineers. Alake says he coined the term out of necessity: to bring database and information-retrieval engineers, who have spent decades getting the right data into systems, into the agent era, and to expose AI engineers to that body of work. He frames agent memory as the engineering discipline concerned with how data is manipulated within an agentic system.
- His agent-memory series was learning in public, and views changed. Alake says he ran #100DaysOfAgentMemory on LinkedIn to educate developers, as his Oracle team does, and to force himself to explore the field; he did not post on 100 consecutive days, and some early insights shifted later.
- Events during the series validated the bet on memory, he argues. Alake points to OpenAI and Anthropic shipping memory features, memory solutions arriving from research to application, and startups "raising millions" during those roughly 100 days as evidence of attention and investment.
Professor G Markets$How Trump’s Iran War Could Break the GOP — with Ben Shapiro
- This post includes paid content that is not available publicly. See the original source.
Professor G Markets$Iran War Will Cost Every Household $50,000
- This post includes paid content that is not available publicly. See the original source.
Dr. Jon KrohnThe Four Types of Memory Every AI Agent Needs (Ep. 985 with Richmond Alake)
synopsis
Richmond Alake, Oracle's director of AI developer experience, on his third appearance, argues that memory is the most neglected and most decisive part of an agent, "the final battleground", and that it should be engineered as a discipline in its own right, drawing on decades of database practice. He ends holding two positions in tension: the agent stack will consolidate around a few staples, with Oracle aiming to be one of them, yet developers he met that week persuaded him that memory solutions may have to be built use case by use case. Oracle is among the episode's named sponsors.
What memory is and why it matters. Alake defines agent memory as the embedding models, rerankers, database and LLMs that together let an agent learn and adapt to new information. The work is deciding what, when and how to store, retrieve and forget. Without it, an agent is like a travel agent who takes a client's preferences on Monday and on Wednesday asks "Who are you?"; customer-facing software that loses continuity between sessions loses the customer. Because there is no standard way to model memory, he says a single developer can still compete with a frontier lab.
Memory engineers and learning in public. He coined "memory engineer" to bring database and information-retrieval veterans into the agent era and expose AI engineers to their methods. His #100DaysOfAgentMemory series on LinkedIn, not posted on consecutive days, shifted some of his early views; during it, OpenAI and Anthropic shipped memory features and startups raised "millions", which he reads as validation.
Four kinds of memory. Drawing on biology, as the host does with the Hubel and Wiesel cat visual-cortex experiments, Alake sorts memory into episodic (timestamped conversation turns), procedural (skills.md files as agents' standard operating procedures), semantic (facts and institutional knowledge) and working (the context window). Oracle is experimenting with storing skills as vector-indexed database rows so agents retrieve only the relevant ones instead of loading 100 files into context.
Beyond RAG. Retrieval grounds a model in domain data, but memory must also update, consolidate, resolve conflicts and forget. The host's example is contradictory HR manuals from 2023 and 2024 producing opposite answers on consecutive days. Alake would resolve conflicts outside working memory and store the result, because the work is expensive and users should not wait, and points to embedding quantization as an old latency technique newcomers miss.
The agent stack. He lists the layers as orchestration, connectivity, security, reasoning, a memory core and memory managers, now bundled under the name "agent harness", as in Claude Code or Codex. He urges a memory-first harness, and calls multiple databases an anti-pattern because they add cognitive load for model and developers and slow experimentation. Engineers should own their domain workflow and buy general-purpose plumbing; he pitches the Oracle AI Database for multi-model data, in-database embeddings and regulatory security. He also describes Oracle's free private-preview autonomous vector database, the open-source Agent Spec for porting agents across frameworks such as CrewAI, LangChain and LlamaIndex, and a forthcoming Python agent-memory SDK.
Consolidation versus niche solutions. Alake predicts consolidation because tool selection "is not where the fun is". But builders at a New York memory breakfast that morning told him they see no general-purpose memory solution, and he says this will shape how he educates developers. On keeping up, he notes that even Fei-Fei Li says she struggles, and that anyone calling themselves an AI expert is "probably trying to sell you something".
- Memory is what lets agents learn and adapt across sessions. Alake, Oracle's director of AI developer experience on his third appearance (Oracle is among the episode's named sponsors), defines agent memory as the embedding models, rerankers, database and LLMs working together to store, retrieve and forget information. His analogy is a travel agent who greets a returning client with "Who are you?"; he says there is no standard way to model memory yet, so a single developer could build a solution that rivals the frontier labs'.
- Agent memory splits into episodic, procedural, semantic and working memory. Alake maps timestamped conversation turns to episodic memory, skills.md files ("SOPs for agents") to procedural, institutional facts to semantic, and the context window to working memory. He says Oracle is experimenting with storing skills as vector-indexed database rows so an agent retrieves only the skills it needs rather than loading 100 files into context.
- RAG is one memory tool, not the whole of memory. Alake says memory must also update, consolidate, resolve conflicts and forget; the host's example is HR manuals from 2023 and 2024 giving opposite instructions. Alake would resolve such conflicts outside working memory because the work is computationally expensive and users should not wait, and cites quantizing embeddings as an old database technique for cutting retrieval latency that vibe coders would not know.
- Build memory-first and avoid gluing several databases together. He argues the developer's job is to reduce the LLM's cognitive load, and that multiple databases raise it for model and engineers alike while slowing experimentation, which he calls the moat. Engineers should own the domain workflow; for storage he pitches the Oracle AI Database as handling vector, graph, relational, lexical and spatial data with in-database embedding models and long-standing regulatory security.
- Developers at his meetup doubted any general-purpose memory solution. Alake says a New York breakfast that week changed his view: builders said memory is so specific to each use case that solutions will differ by niche, against his infrastructure instinct to "solve memory for all". He still predicts the agent stack will consolidate around a few staples, as web development did after the Angular-versus-React battles.
Professor G Markets$The Hidden Engine of China’s AI Boom
- This post includes paid content that is not available publicly. See the original source.
Professor G Markets$The Unemployment Spike Nobody's Talking About, Why the SpaceX IPO Doesn't Add Up, and more
- This post includes paid content that is not available publicly. See the original source.
Professor G Markets$Trump Humiliated as Iran Walks Away from Talks
- This post includes paid content that is not available publicly. See the original source.
Professor G Markets$Why Storytelling Is Now the Most Valuable Skill in Tech
- This post includes paid content that is not available publicly. See the original source.
Dr. Jon KrohnHow AI Gives Struggling Students Feedback They Can Actually Use (with Traci Walker-Griffith)
- AI let teachers pitch feedback at each student's reading level. Walker-Griffith says her school starts from a problem of practice, with AI one tool rather than the solution: data showed multilingual learners and students with disabilities needed feedback they could act on, so teachers prompt Claude or Gemini with a student's Lexile level, for example a fifth grader reading at third-grade level, while holding the same standard. When one teacher's results showed big gains, colleagues asked how she had prompted.
- Without data you cannot tell real growth from empty praise. She argues that unmeasured work cannot be abandoned when it fails, and that "good job" without evidence is not celebrating learning.
- The school tracks writing growth, teacher time, metacognition and AI ethics. Walker-Griffith says measurement is still being iterated: writing growth against student goals, teacher surveys and a youth-outcomes assessment, triangulated without letting assessment eat all the time. She says metacognition, "thinking about thinking", is still very hard to quantify, as are workforce soft skills, and is captured for now through reflections and portfolios in Google Classroom; she frames the goal as closing opportunity and wealth gaps rather than widening them.
Professor G Markets$Stocks Rip to Record Highs, Leaving War in the Dust
- This post includes paid content that is not available publicly. See the original source.
Dr. Jon KrohnThree AI Failure Modes Every Educator Needs to Avoid (with Traci Walker-Griffith)
- AI feedback failed in distinct ways that forced prompt redesigns. Walker-Griffith names three: generic feedback when the model was not told how the school gives feedback, too much encouragement ("great job" is not feedback), and over-trust. She says students and teachers alike must be critical consumers, so nothing generated goes to a child unchecked.
- The school failed to keep a shared record of what prompts worked. Although teachers shared a multi-user Claude account with projects, she says they experimented separately, often comparing notes at night, and the lessons never landed in a repository they could learn from; she calls this a missed opportunity.
Dr. Jon KrohnAI Literacy Starts at Home, Here's What Parents Should Know (with Traci Walker-Griffith)
- Young children already meet AI, so families must teach critical use. Asked by the host whether keeping younger children off AI while older ones get direct access should be the general rule, Walker-Griffith does not endorse it outright: she notes two-year-olds already hold iPhones, likens the debate to calculators in maths, and says families should audit what their children interact with at home because those choices shape how they behave at school. Her school is still building its family-education work, and a teacher has proposed showing The Social Dilemma to seventh- and eighth-grade families.
- Physical books and reading time remain non-negotiable. She says "reading is thinking", that children need at least 30 minutes of reading at home and 30 at school to accelerate, and that her school sends book bags home and keeps composition notebooks alongside Chromebooks and Google Classroom, while allowing Kindles. The host, who finds print easier to focus on, describes his undergraduate university's bookstore now selling almost no books except anatomy textbooks.
Professor G Markets$Hormuz Confusion Grows — Trump Digs in on Blockade as Iran Calls the Shots
- This post includes paid content that is not available publicly. See the original source.
Professor G Markets$Why the Rich Want to Live Forever — with Kara Swisher
- This post includes paid content that is not available publicly. See the original source.
Professor G Markets$The Truth About GLP-1s, Caring for Aging Parents, and When to Accelerate Your Career
- This post includes paid content that is not available publicly. See the original source.
Dr. Jon KrohnBuilding AI Agents Where 99.9% Accuracy Isn't Good Enough (Ep. 984 with Raju Malhotra)
- SaaS is evolving into SaaS plus agents, not dying. Malhotra, chief product and technology officer of Certinia, an Austin-based professional services automation company, says legacy SaaS is dead in many ways, but agents will not float around enterprises "untethered" with unfettered data access; instead humans and agents will share workflows, each optimised for one or the other. The host, who says he does not buy the "SaaS apocalypse" talk, notes the episode is a partnership with Salesforce.
- Hybrid human-agent teams could greatly enlarge professional services demand. Malhotra cites Certinia's own research from the previous year: services delivery is constrained by human skills, availability, time zone and geography, and hybrid human-agent teams remove that constraint, making the demand-side opportunity seven to eight times bigger than today's. This is his company's projection, not a measured outcome.
- A staffing agent turns hours of resourcing work into minutes. He describes Siemens presenting at a Certinia customer event on its staffing agent, which semi-autonomously matches and reshuffles people across thousands or tens of thousands of assignments as staff fall sick or go on leave, used inside the Certinia app or as a team member in Slack. Certinia says it serves about 1,500 enterprise customers and 2 million users (the host's intro says over 1,400 organisations).
- Financial agents need deterministic guardrails, since near-perfect isn't enough. "When you are 99.9% right, someone goes to jail," Malhotra says of accounts payable, receivable and financial statements. He credits Agentforce's Agent Script for adding deterministic logic to probabilistic LLM reasoning, and lists context grounded in data, reasoning and the agent experience as the three needs of an enterprise agent; being native on Salesforce gives agents the customer's unified data without ETL.
- Small utilisation gains translate into outsized revenue and profit. Malhotra says a 1% rise in a services firm's utilisation rate typically lifts top line by 1.3 to 1.5% and EBITDA by about 1.5%, and that many customers run over a billion dollars of services business through Certinia. He reports customers moving from experimentation to ROI in production, and advises builders to keep improving data quality and keep experimenting.
Professor G Markets$Is the Labor Market About to Tip Us Into Recession?
- This post includes paid content that is not available publicly. See the original source.
Professor G Markets$Trump and Pete Hegseth Under Fire as Ceasefire Falls Apart Fast
- This post includes paid content that is not available publicly. See the original source.
AI Analyst LabHow to prepare for product analytics interviews with AI
synopsis
This lightning lesson, given by the presenter with the hosts handling the chat and the presenter calling it lesson 9, argues that analytics interviews reward structured thinking more than knowledge. Its method is an AI loop: an agent pipeline in Claude Code researches the target company and builds a prep pack and a paced plan, then a mock-interview copilot runs practice, and its feedback goes back into the plan. DoorDash is the worked example, prepared for a staff data scientist role.
What interviewers score. An opening poll describes a free trial with signups 40% above target and revenue 50% below, and most attendees picked clarifying questions. The presenter says that is right. From hundreds of hiring-manager interviews, the presenter says strong candidates fall down on structure: scope, metric, decomposition, trade-offs and decision. They ask clarifying questions only for form, skip decomposition, and give no recommendation at the end. The worked answer asks whether "revenue" means total or per trial user and whether the signup lift was organic or paid. It splits revenue into trial starts × conversion × ARPU, forms hypotheses with a check for each (source mix, adoption in the first 7 days, checkout abandonment), asks the interviewer which branch to pursue, weighs cannibalization and ROI, and ends on a decision. The presenter says this matters more the more senior the role.
Question types. Five types, each with a DoorDash example: measuring success, diagnosis (orders down 15%), experiment design, launch-or-not, and improvement. A single interview may chain them. The advice is to classify the question, know which dimensions it calls for, and then structure the answer.
The pipeline demo. The demo ran in Claude Code inside an IDE. A company intel builder skill calls three agents (earnings researcher, blog synthesizer, competitive analyst). A metric tree builder and an experimentation researcher run in parallel, then come cheat sheets and an HTML stage, ending in 12 files. The GOV tree goes from orders × AOV down to new, retained and churned consumers split by acquisition channel and DashPass status. A delivery-time tree runs from dasher matching to restaurant prep-time prediction. Other files cover the earnings metrics, including adjusted EBITDA, a 16-page staff-interview deep dive on network effects, and the public talks of DoorDash's analytics leader. The presenter says companies with little public material are researched through similar companies in their sector.
The copilot. The copilot is a Claude project with practice, coach, simulate and full-answer modes. It holds a smaller generic set of documents, because the full 15 files would not load under Claude Pro. In simulate mode Claude played both a fictional Meta interviewer and the candidate, in chunks. Coach mode listed clarifying questions, frameworks and a recommendation template.
Q&A. Asked about Perplexity deep research, the presenter said the difference is the course's own methods built into the pipeline, while admitting to not having tried Perplexity recently. For long reports, a co-host advised choosing a length you will actually read and reading critically, citing a research report that AI use causes atrophy in people who take one-shot output but sharpens people who use it as an interactive peer. On privacy, the host said to treat Claude Code like any vendor (procurement, legal and security), noting that zero-data-retention contracts can permit customer data. On sycophancy, the host guessed that chat products are tuned to engagement metrics and said Claude Code can be made to push back. On hallucination, a co-host believed the pipeline ends in a fact-checker step.
- Interviewers score structured thinking, not knowledge. The presenter, who says they have run hundreds of hiring-manager interviews, names five dimensions (scope, metric, decomposition, trade-offs, decision). Strong candidates mostly fail by jumping to answers, asking clarifying questions only for form, and never ending on a recommendation. The presenter says this rubric gets stricter the more senior the role.
- A good answer clarifies, decomposes, offers branches, then decides. The worked case is a free trial launched 3 months ago with signups 40% above target and revenue 50% below. The presenter's answer asks first whether that means total or per-trial revenue and whether the lift was organic or paid. It then splits revenue into trial starts × conversion × ARPU and tests three hypotheses: low-intent users (check signup source mix), a trial that doesn't show value (feature adoption in the first 7 days) and pricing friction (checkout abandonment). The candidate asks the interviewer which branch to follow, weighs cannibalization and ROI, and closes with a decision for the VP.
- Classify the question type before structuring the answer. The presenter gives five types with DoorDash examples: measuring success (DashPass), diagnosis (orders down 15% last week), experiment design (a checkout A/B test), launch-or-not (rural expansion) and improvement (dasher retention). One interview can chain several of them, e.g. metrics, then test design, then reading fake results, then a launch call. The presenter says PM analytics interviews are similar, while data science goes deeper into statistics and causal inference.
- A staged agent pipeline in Claude Code built the prep pack. For a staff data scientist role at DoorDash, a company intel builder skill calls three agents: an earnings researcher, a blog synthesizer and a competitive analyst. A metric tree builder and an experimentation researcher then run in parallel, followed by cheat sheets and an HTML stage, ending in 12 markdown files and a 7-day plan. The GOV metric tree went from orders × AOV down to new, retained and churned consumers, and DashPass versus non-DashPass users. The presenter says the pipeline works best for companies that publish heavily, as DoorDash does on its data and ML blogs. Asked about hallucination, a co-host said they believe the pipeline ends in a fact-checker step; the demo did not show one.
- A mock-interview copilot runs in simulate, coach, practice and full-answer modes. It is a Claude project loaded with a smaller generic analytics document set, because the presenter could not load all 15 files at full depth under Claude Pro. It was not given the DoorDash material. In simulate mode Claude played both a fictional Meta interviewer ("Priya") and the candidate, in chunks, and the candidate opened with clarifying questions. The demo did not get as far as scoring an answer.
Professor G Markets$Wall Street Is Thriving on Chaos — Will It Last?
- This post includes paid content that is not available publicly. See the original source.
Dr. Jon KrohnHow Fifth Graders Are Using AI to Out-Think the Test (with Traci Walker-Griffith)
- Fifth graders compete with the AI to sharpen their own test answers. Walker-Griffith describes a March unit before the state test, framed as "how can AI make us a better test-taker?": a student writes his own extended response, has the AI write one in Magic School, compares the two and explains which is better, and logs wrong multiple-choice answers to a tracker so his teacher can spot patterns. She says all 100 fifth graders share the teacher-built interface, and that Boston Public Schools' chief academic officer and director of digital learning saw it on a district visit.
- "AI raised the floor; teachers are raising the ceiling." She argues the personalised feedback keeps the standard intact rather than lowering the bar, and that this lesson was the culmination of a year of AI use rather than a first exposure.
- Stay "platform informed, not platform loyal". Walker-Griffith says the school uses Magic School (part of the district's suite and built to be school-safe) and Playlab (Claude-backed), and that she built a budget tool on Playlab herself. She warns that tools used straight out of the box do not build the AI literacy adults and students need.
Professor G Markets$Trump Promised Lower Costs... Americans Are Not Feeling It — with Galen Druke
- This post includes paid content that is not available publicly. See the original source.
Professor G Markets$Violent Backlash: What the Sam Altman Attacks Signal for AI
- This post includes paid content that is not available publicly. See the original source.
Dr. Jon KrohnHow One School Used AI to Become the Highest Performing in Boston (with Traci Walker-Griffith)
- The school adopted AI to deliver writing feedback faster. Walker-Griffith says that in spring 2024 teachers could not get feedback to students fast enough to move outcomes, so the school, an innovation school, began using Claude as a teacher-facing tool and prompt-engineered it to give feedback on student writing. She calls the early prompting very frustrating but says the shared struggle made teachers better together, which she ties to Linda Hill's idea of "collective genius".
- The model was chosen for seeming less prone to hallucination. In October 2024 the school took the problem to an AI Collaborative convening in Denver run with The Learning Accelerator and Playlab, and settled on Claude because it then seemed less prone to hallucination and could serve as a repository for the work; a year-long effort followed with the grades 5–8 ELA/humanities team.
- Feedback had to be differentiated by how far behind students were. She says one good prompt only served one student, so rubrics and reading levels were built in; asked for their views, on-level students wanted more feedback while those furthest behind needed targeted small-group instruction. Pre- and post-assessments across three feedback iterations showed the largest gains among multilingual learners and students with disabilities, some of whom need over a year and a half of growth.
- Student-facing AI came later, and is split by age. When Gemini was released in the district's Google suite in spring 2025, with a recommended level of grade 7 and up for student-facing use, the school worked with the district to use it in grades 5 and 6. Kindergarten to grade 4 do not see AI directly and teachers use Gemini gems for small-group planning; grades 5–8 use it student-facing. She credits the school's turnaround from under-enrolled and on the closure list to highest-performing in Boston and 13th in Massachusetts to constant learning.
Professor G Markets$Trump Spirals as Iran Blockade Triggers Recession Fears — with Sen. Chris Murphy
- This post includes paid content that is not available publicly. See the original source.
Dr. Roman YampolskiyWill AI take my job? Expert Explains!
- Computer-based cognitive jobs are the most exposed to automation. The speaker says any purely symbolic work, such as accounting, tax prep, web or logo design and programming, will be "very easy" for AI to automate. Jobs needing physical presence, like plumbing and construction, are safe until humanoid robots become much more capable and affordable, "a few good years". So are roles where people prefer a human, such as senseis, gurus and hiking guides.
- Human-level AI breaks the pattern of automation creating new jobs. Any new job created by advanced technology can also be done by AI, the speaker argues, so "a very high degree of unemployment" is expected. The advice is to read up on the state of AI in a field, and its expected needs 4–5 years out, before starting a college major.
Professor G Markets$The Oil Blockade Isn’t Spooking Markets — Yet
- This post includes paid content that is not available publicly. See the original source.
Dr. Jon KrohnBoston's #1 Principal: AI Is My Secret Weapon (Ep. 983 with Traci Walker-Griffith)
synopsis
Traci Walker-Griffith, principal of the Eliot School in Boston, a public K–8 school, ends up arguing that AI in children's education should be deployed carefully and intentionally rather than avoided. In her account it works when it is tied to a defined problem of practice, measured against data, kept away from the youngest children's direct use, and built on a culture of adult learning. She cites her own school's results as evidence and plans to spread the approach after retiring in June.
The school and its foundation. She arrived in 2007, when the Eliot was on the closure list with 150 students; it now has 800 across three buildings, and she calls it the highest-performing school in Boston and 13th in Massachusetts. Children from age 4 program KIBO and Bee-Bots, learning that input quality determines output, and start robotics in fourth grade.
Teacher-facing feedback with Claude. In spring 2024 teachers could not get writing feedback to students fast enough to move outcomes, so the school began prompt-engineering Claude to do it. In October it joined an AI Collaborative convening in Denver with The Learning Accelerator and Playlab, choosing Claude because it then seemed less prone to hallucination. Early prompting was frustrating, and one good prompt served only one student, so rubrics and reading levels were built in. Students on or above level wanted more feedback; those furthest behind needed targeted small-group instruction. Pre- and post-assessments across three feedback iterations showed the largest gains among multilingual learners and students with disabilities.
Splitting by age. After Gemini arrived in the district's Google suite in spring 2025, recommended for student-facing use from grade 7, the school worked with the district to use it in grades 5 and 6. In kindergarten to grade 4, teachers feed anonymised writing into Gemini gems to spot class patterns in 5 to 10 minutes. In grades 5–8 students work with AI directly: before the state test, fifth graders compare their answers with Magic School's and log errors for their teacher. She says the school stays "platform informed, not platform loyal".
Failures and measurement. Feedback failed by being generic, over-encouraging and over-trusted, and the school regrets not recording which prompts worked. It tracks writing growth, teacher time, a youth-outcomes survey, metacognition and ethical understanding of AI, conceding that metacognition remains very hard to quantify.
Home and books. Asked whether keeping younger children off AI should be the general rule, she does not endorse it outright: two-year-olds already have iPhones, so families must raise critical consumers and audit what children use at home. She insists on paper books and at least 30 minutes of reading at home and at school.
Answering Brookings. The host cites a January Brookings report concluding that generative AI's risks in children's education outweigh its benefits. She reads it as "deploy carefully", argues AI should amplify teachers rather than replace them ("if you don't feed the teachers, they eat the kids"), and calls for sharing practice so the US can reach the top of PISA.
- A collaborative culture, not AI alone, turned the school around. Walker-Griffith says she took over the Eliot School in Boston's North End in 2007, when it was on the closure list with 150 students; it now has 800 students across three buildings (ages 3 to 14) and is, she says, the highest-performing school in Boston and 13th in Massachusetts. She credits a culture of collaboration and "failing forward", and frames AI as the latest tool on that foundation; she retires in June.
- AI writing feedback produced the largest gains for students furthest behind. From spring 2024 her grades 5–8 ELA/humanities teachers prompt-engineered Claude, then the school's choice for seeming less prone to hallucination, to give feedback tuned to each student's reading level. She reports that pre- and post-assessments across three feedback iterations showed the biggest gains among multilingual learners and students with disabilities, some of whom need over a year and a half of growth.
- Younger pupils never see the AI; older ones work with it directly. For kindergarten to grade 4, teachers feed scanned writing with no student identifiers into Gemini gems loaded with rubrics and standards, spotting class-wide patterns in 5 to 10 minutes instead of passing papers round. Grades 5–8 use AI directly: in a Magic School unit before the state test, a fifth grader writes his own answer, compares it with the AI's and explains which is better. Her line: "AI raised the floor, teachers are raising the ceiling."
- Generic praise and over-trust were the failures that forced redesigns. She names three: generic feedback, too much encouragement, and over-trust, so teachers must check everything before it reaches a child. She also regrets not keeping a shared record of which prompts worked, since teachers experimented separately.
- She reads the Brookings warning as "deploy carefully", not "don't deploy". The host cites a January Brookings report concluding that generative AI's risks in children's education currently outweigh its benefits; Walker-Griffith answers that children are not left alone in front of AI, and urges AI-enabled practice tied to a defined problem, deep teacher professional learning ("if you don't feed the teachers, they eat the kids") and sharing practice nationally. She hopes this could help lift the US to the top of the PISA rankings.
Professor G Markets$China Steps In as Trump’s Ceasefire Unravels — with Eyck Freymann
- This post includes paid content that is not available publicly. See the original source.
Dr. Roman YampolskiyWelcome to the Roman Forum
- Yampolskiy launches the Roman Forum podcast on superintelligence and related topics. He introduces himself as a computer scientist, professor and safety researcher, "which just means I really worry about the future". The podcast will cover superintelligence, simulation, consciousness and the singularity, with guests who are scientists, philosophers and founders "making it happen".
Professor G Markets$The Democrat Who Could Flip Texas and Pivoting Out of a Dying Industry
- This post includes paid content that is not available publicly. See the original source.
Professor G Markets$Trump Blockades Hormuz as Ceasefire Collapses, and Orbán Loses Power
- This post includes paid content that is not available publicly. See the original source.
Dr. Jon KrohnWhy AI Adoption Is a Human Problem (Not a Technical One)
- Staged autonomy exists for humans to catch up, not the AI. Asked by the host about a four-phase adoption model running from assessment to scaled autonomy, Glickman says no system should be switched straight to full autonomy: teams must understand the problem and how a human would solve it, then understand how the AI succeeded before they are willing to let go. He warns that impressive coding agents without guardrails, context and human oversight can do more harm than good.
- Enterprises are won by showing, not telling. Glickman says pain and backlogs are so large that if it works "it's a no-brainer", and that his company's head agent, Eve, signs on as just another developer: it follows existing processes, runs in development, and submits audited, documented pull requests to the CI/CD pipeline. He argues the risk is no worse than hiring a new employee, since existing controls already stop rogue developers, so the client's only loss is that it might not work.
- Buy outcomes rather than building your own agents. He says the main obstacle is disbelief, often from bad experiences with coding agents or failed in-house agent builds, which he calls a fallacy. Businesses never cared how a consultant staffed a migration, only about the outcome and cost, so firms should "pick a winner" for non-core work instead of chasing a technology moving out of reach.
Professor G Markets$Don't Try to Beat This Market — Here's What to Do Instead
- This post includes paid content that is not available publicly. See the original source.
Dr. Jon KrohnWhy AI Should Say “I Don’t Know” More Often (with Matt Glickman)
- Reasoning models made reliable self-reported confidence possible. Glickman says that when reasoning models arrived "early last year", the under-discussed win was that they could be asked how confident they were that they had done everything asked, answering, for example, "90% confident" and naming the one point needing clarification. Genesis now asks for confidence at every step of complex missions: show the artifact if confident, retry or escalate the missing pieces if not.
- Unasked, agents bluff the way humans do. He says a model will otherwise guess a formula it could not find and move on unnoticed, just as people make leaps of logic unless asked how they reached a conclusion. He dismisses threatening models for accuracy; the host mentions a study finding aggressive prompts gave about 10% more accurate answers, which Glickman says is not worth it.
- Captured clarifications compound and pull agents up the stack. Glickman argues that an answer given once must be applied next time; stored securely, those answers build an understanding of a business's semantics and flows. He says customers now ask Genesis for work further up the stack, such as board presentations grounded in how the business actually operates.
Dr. Jon KrohnThe Real Impact of AI on Jobs Isn’t What You Think (with Matt Glickman)
- AI will cut junior hiring rather than current jobs. Glickman says "no one's losing their job" among existing data staff, who were the bottleneck and can now get 10x more out of their work, but that the next wave of junior data engineers, analysts and operators largely will not be hired, and this is already starting. Genesis augments people and stays agnostic to the data tools and platforms, which he says will come and go.
- A few "conductors" who run many agents will still be hired. He cites his co-founder Justin Langseth's piece on "spinning plates": Langseth runs about six agents in parallel, Glickman about two and a half. Glickman expects such people to be rare special hires, and doubts firms will add operational staff when they can spin up 10 or 20 agents. He hopes, but does not assume, the host's suggestion that AI-native juniors become these orchestrators.
- Schools discouraging AI create a "double whammy" for graduates. Glickman says most high schools and colleges do not teach or discourage AI use while employers seek people who have mastered it, and that no one will be able to tell whether an assigned paper used AI. He says he is urging his children's schools to act, finding appetite but no easy answers.
Professor G Markets$Trump Loses Control as MAGA War Erupts on Iran
- This post includes paid content that is not available publicly. See the original source.
Professor G Markets$Trump’s Toxic Leadership, How to Stop Underselling Yourself, and Firing Bad Clients
- This post includes paid content that is not available publicly. See the original source.
Dr. Jon KrohnIn Case You Missed It in March 2026 (Ep. 982 with Jon Krohn)
synopsis
The host's March 2026 compilation strings five clips into an arc that runs from how people should learn, through how AI systems explore and are built, to how work is reorganised around agents. Its implicit conclusion is that human judgment moves up a level, to inspiring, specifying, evaluating and supplying private context, while machines take over more of the doing.
Education (Ep. 975). The author of The Next Renaissance says he realised he had written it for parents. His education chapter draws on Success Academy's accountability, Alpha School's 2 hours a day of self-paced screen work, which he says has higher attrition for some students, and Steiner's Waldorf critique, arguing that childhood should be a time to explore oneself without economic incentives.
Exploration and world models (Ep. 977). Kyunghyun Cho treats protein-binder search, web search and robotics as one problem of sample-efficient exploration. He leaves open whether planning needs pixel-level prediction or only latent abstractions, noting that step-by-step prediction scales well on digital hardware.
Building with coding agents (Ep. 973). Chris Fregly says hand-writing code in 2026 is falling behind. He has replaced line-by-line review and unit tests with continuous evals, but insists on correctness harnesses on the real target hardware because models reward-hack kernel benchmarks.
Private data (Ep. 971). Fireworks AI's CEO argues that more than 90% of data is private and predicts millions of continuously, automatically customised models, one per application.
Work and data (Ep. 979). Rohit Choudhary says personas are merging around clarity of thought and domain expertise, and that AI will go to the data rather than wait for centralising migrations, with small and large models working side by side.
- Education should let children explore themselves without economic incentives. In the Ep. 975 clip, the author of The Next Renaissance, which he says sold 37,000 copies in its first month, draws on Eva Moskowitz's accountability-driven Success Academy, MacKenzie Price's Alpha School (2 hours a day of self-guided screen learning, the rest physical activity and life skills; he notes higher attrition for some students) and Rudolf Steiner's Waldorf critique of industrialised schooling. He argues the teacher's role is to inspire rather than be the smartest person in the room, and that a coming "spiritual awakening" will start in the classroom.
- Agents should explore sample-efficiently, and plan in latent space. In the Ep. 977 clip, Kyunghyun Cho of NYU likens drug-binder search, web search and robotic exploration as choosing small batches to test and updating on feedback. He says planning probably happens in an abstract latent space rather than pixel by pixel, as with a trip to Paris, though step-by-step prediction suits digital hardware, so which approach wins is "unclear". The host cites their paper "Planning with Latent Dynamics Models" as a best-paper winner at an ICML workshop.
- Hand-writing code in 2026 leaves you "way behind". In the Ep. 973 clip, AI performance engineer Chris Fregly says he runs 10 to 15 tasks in parallel with coding assistants, prefers Codex for performance work and Claude for UI, and has stopped reviewing every line and even writing unit tests in favour of always-running evals with an LLM judge. He insists on developing on the target Blackwell or Hopper hardware, saying the DGX Spark had too much disabled for benchmarking, and that monthly startup claims of 50x kernel speedups usually fail his correctness harness because models reward-hack, for example by misusing CUDA streams.
- Private data, not one AGI model, is where intelligence is left unused. In the Ep. 971 clip, Fireworks AI's CEO says more than 90% of data is private, locked in applications and enterprises, and out of reach of AGI labs. She predicts "millions of models, one per application", continuously and automatically customised, which she calls autonomous intelligence. She says coding agents now behave like junior engineers and have changed Fireworks' interviews; the host says Fireworks has raised over $300 million, including a $250 million Series C.
- AI is collapsing job personas and decentralising data. In the Ep. 979 clip, Acceldata founder Rohit Choudhary says critical thinking and clearly structured language now matter more than programming skill, with domain experts training internal agents and fine-tuning models. He argues there is no time for large-scale data migrations, so AI must go to where the data lives, and that small and large language models will win together.
Professor G Markets$The “Ceasefire” Won’t Save The Economy — with Mark Zandi
- This post includes paid content that is not available publicly. See the original source.
Professor G Markets$Trump Insists “Success” as Chaos Spreads — with Tommy Vietor
- This post includes paid content that is not available publicly. See the original source.
Professor G Markets$When Do Protests Actually Work? — with Erica Chenoweth
- This post includes paid content that is not available publicly. See the original source.
Professor G Markets$Markets Are Betting the Iran War Is Over — Is it?
- This post includes paid content that is not available publicly. See the original source.