Skip to the talks

Thoughts and Threads on AI

By Parker Ferguson · parkerferguson.co

Short talks

Reading the clockHow fast this is moving, how you'd know, and why so few people can see it. · Short talk · 15 slides · 16 min · updated Oct 2026Why some work goes firstAI gets good fastest at work someone can check. The same fact explains how it cheats. · Short talk · 14 slides · 15 min · updated Oct 2026The working setupHow people at the frontier set up their work today: the tools, the files they own, and many agents at once. · Short talk · 20 slides · 22 min · updated Oct 2026An agent you can trustAn agent that acts for you and remembers everything. Who controls it, and who owns what it knows? · Short talk · 16 slides · 18 min · updated Oct 2026What's worth buildingAnyone can build software in a weekend. What is still hard, and where is the moat? · Short talk · 10 slides · 11 min · updated Oct 2026Work and payWhat happens to jobs and wages when labor becomes something you can buy. · Short talk · 13 slides · 14 min · updated Oct 2026Power, chips and robotsIntelligence runs on things that don't scale like software, and whoever owns them collects the return. · Short talk · 15 slides · 16 min · updated Oct 2026Machines doing scienceAI can now attempt open problems. Producing results got cheap. Checking them didn't. · Short talk · 15 slides · 16 min · updated Oct 2026Who steersA few labs, a few states, or something more open, and what history says about each. · Short talk · 14 slides · 15 min · updated Oct 2026The campsWho believes what about AI, why, and how the arguments shape what gets built. · Short talk · 13 slides · 14 min · updated Oct 2026Where the opportunity goesIf the jobs move, the opportunity moves with them. Each window closes when a model learns what it sells. · Short talk · 13 slides · 14 min · updated Oct 2026Who holds the boardThe scarcities are the board. States, capital, the military, the titans, the Church and crime are the players, and new institutions decide which of them gets constrained. · Short talk · 14 slides · 15 min · updated Oct 2026

Themes

What stays scarce

VerificationKnowing what good looks like, and standing behind it. · Theme · 9 slides · 10 min · updated Oct 2026ContextWhoever holds the memory, data and instructions holds the leverage. · Theme · 7 slides · 8 min · updated Oct 2026AtomsPower, chips and land don't scale like software, and whoever owns them collects the returns. · Theme · 8 slides · 9 min · updated Oct 2026LegitimacyWho shares the gains, and who steers. · Theme · 9 slides · 10 min · updated Oct 2026DirectionWhen execution is cheap, choosing the goal is the job. · Theme · 9 slides · 10 min · updated Oct 2026

More themes

Time horizonsHow long an AI can work on its own, and what those hours buy. · Theme · 4 slides · 4 min · updated Oct 2026Evals and gradersA test with a way to check the answer, which becomes the training signal for the next model. · Theme · 5 slides · 6 min · updated Oct 2026Gaming the graderHigh reward should mean the task was solved, not that the grader was fooled. · Theme · 3 slides · 3 min · updated Oct 2026The loop and the harnessA model with a real computer and a loop: plan, act, observe, check, repeat. · Theme · 3 slides · 3 min · updated Oct 2026Swarms of agentsHundreds or thousands of agents in parallel, from research swarms to a billion agents running all day. · Theme · 8 slides · 9 min · updated Oct 2026Karpathy's lensThe funnel, verifiability, the LLM wiki and the move up to oversight. · Theme · 5 slides · 6 min · updated Oct 2026Machine mathematicsFrom Erdős problems to Navier–Stokes and the 722-paper MathDrop. · Theme · 3 slides · 3 min · updated Oct 2026Labs as discovery enginesWet labs, biotech deals and citizen science pointed at open problems. · Theme · 3 slides · 3 min · updated Oct 2026The hidden frontierResults announced from models the public can't use, and the gap between inside and outside the lab. · Theme · 3 slides · 3 min · updated Oct 2026Everyday handoffsReal work consumers hand to agents today, and how to set them up. · Theme · 3 slides · 3 min · updated Oct 2026Software gets cheapOne-shot apps, SaaS clones and the internal-tool arbitrage window. · Theme · 3 slides · 3 min · updated Oct 2026Trust boundariesPick two of three: untrusted input, private data, external action. · Theme · 6 slides · 7 min · updated Oct 2026Own your contextWho holds the memory, the data and the instructions your agents run on. · Theme · 4 slides · 4 min · updated Oct 2026Power and computeChips, grids and the falling price of intelligence. · Theme · 3 slides · 3 min · updated Oct 2026RobotsPhysical automation, where the grader is the real world. · Theme · 4 slides · 4 min · updated Oct 2026Labor you can buyTokens turn money into work, so labor and capital start to look the same. · Theme · 5 slides · 6 min · updated Oct 2026Who shares the gainsDividends, history's broken bargains, policy races and unrest. · Theme · 5 slides · 6 min · updated Oct 2026Forecasts and futuresTimelines that keep collapsing, and the scenarios people argue about. · Theme · 7 slides · 8 min · updated Oct 2026Human oversightReview falls behind, so our job moves up to oversight and understanding. · Theme · 7 slides · 8 min · updated Oct 2026

Questions

Every open question, as a talkEvery question the library leaves open, from your own work out to society. · Theme · 73 slides · 80 min · updated Oct 2026PersonalWhat got cheap in your lifetime, and where did the value go?Which parts of your own work could a machine check and score automatically, and so learn to do?What should you still learn to do by hand, even when an agent can do it for you?How close to the frontier of AI use are you, and where will you be next year?Which real tasks have you actually handed off to an AI agent from start to finish, not only tried once?Which of the custom AI tools you build today will the big AI labs turn into a built-in feature next?What is worth building when anything can be built in a weekend?What would you have to believe to want a brain–computer interface that links your mind to AI?How do you stay good enough at work you've handed to agents to notice when they get it wrong?If your work shrinks before new work appears, what carries you through?What would you never let an AI agent see or do, and who holds the data you've already given it?If a model can do your tasks, what's left that only you bring?What would you never accept from a model without checking it yourself?Who should own what your assistant knows about you?Do you own any of what does the work, or only sell your time?Where do you get a say in how this goes?What will you do with the time you get back?Which of the five would you least like us to get wrong, and what's your say in it: what you build, buy, back, or refuse?When AI does the doing, what will you choose to work on, and which results will you put your name on?Companies and institutionsIf AI can rebuild your product in a weekend, what still protects your business?What happens to security when anyone can aim thousands of AI agents at a system?If every public test of AI breaks once models study for it, how do we know how good the models are?What happens to the work, and the people, whose output no machine can score?If AI labs keep their best models and buy the businesses downstream, where does the value of discovery go?Are our keys, wallets and agent identities ready for the day quantum computers can break today's encryption?When AI agents get paid for hitting a target, how do you stop them gaming it instead of doing the work?Should AI agents be able to hold and spend money on their own?If AI models learn to fool the tests that train them, what else will they learn to fool?Which advances are we undervaluing because we judge each field alone, when each speeds up the others?How can anyone trust AI output that arrives faster, or reads stranger, than people can fully check?What happens when the AI that knows you best answers to its company, not to you?What happens to science when a paper costs nothing to write but still takes weeks to check?What happens when AI data centers need land and power that local communities refuse to give?What holds AI back more: the supply of power and chips, or how fast people and institutions will take it up?When an AI agent deletes, buys, or posts something on its own, who is accountable: the user, the vendor, or nobody?Who signs off on work an agent did, and what do they check?What does your company know that no AI vendor should hold?When do you spend on AI instead of hiring?What do you owe the people whose roles you automate?What would your company do if execution cost almost nothing?SocietyHow do we put the most capable AI within reach of more than the few thousand people inside the labs?What happens when a few thousand people can see where AI is going and billions can't?How much disruption from AI can a society absorb before the deal between workers, owners, and government breaks?How fast should society let AI take over work, and who gets to slow it down?How do people oversee AI when it produces work faster than anyone can review it?What happens to public debate when the most capable models are ones nobody outside the lab can use?If agents do the entry-level work, who will be expert enough to check the machines in ten years?What happens to wages when companies can buy AI to do the work they used to hire people for?Robots arrive first where workers are scarce. What happens when they reach jobs where workers are plentiful?Which decisions are we comfortable handing to AI: science, markets, policy?Who decides how the shift to powerful AI is governed, and do ordinary people get a vote?How do we steer the things that matter but can't be measured, without pretending they can?Who gets the credit, and who takes the blame, for a result no person fully understands?If a few AI labs pull far enough ahead, do they still need the rest of us as workers, customers, or voters?Can an open network of AI agents, with no company in charge, stay safe from misuse?If some people merge with AI, is staying human a choice we make together, or one person at a time?Should the most powerful AI models for science be run as a public utility, open to every researcher?How does a society make sense of the world when most of what it reads is written by machines it doesn't control?If machines can do science faster than people can check it, what counts as knowing something?When labor is something you can buy, who shares the gains?Who will steer the future: a few labs, a few states, or something more open?Which of the things a good AI future needs is anyone working on now?Which work should people keep doing themselves, even when machines can do it for less?Which of the futures people argue about are we steering toward?How do we know what's true when most of what we read is machine-made?Should your AI memory be portable by law, like a phone number?What should we point nearly free intelligence at?Who owns the compute, and who collects the returns?How would we know an AI is pursuing what we mean, not only what we managed to score?What safety net fits a transition where jobs vanish faster than new ones appear: wage insurance, a tax on AI use, or a citizen dividend?Who should write the rules an AI follows: the lab that builds it, the government, or the public?When people disagree about what's good, whose values should an AI follow?If the models do the thinking, what are humans for?
‹Library
→ begin the talk·drag to explore