Pete Soderling✓

I’m hiring a Chief of Staff to help run @ZeroPrimeVC & @AICouncilConf. It's a mini-COO role for an unusually capable generalist. Most of your time will go to operating and growing AI Council. The rest goes to increasing my leverage across both AI Council and Zero Prime: turning priorities into action, preparing decisions, protecting focus, and making sure important threads don't drop. The right person is a high-agency generalist - comfortable owning a P&L, fully AI-pilled in how you work, and low ego about the job. You should move comfortably between commercial work, marketing, events, finance, executive communication, and operational detail. The scope is intentionally broad and expected to grow. The right person will earn ownership quickly and can develop toward broader operating leadership over time. Evidence of ownership matters more to me than a matched resume. $150–180K base, bonus, carry eligibility. Full time, in person, San Francisco. To apply, send resume and LinkedIn profile to community@zeroprime.vc, with short answers to two questions: 1. What's the most ambiguous cross-functional problem you've personally owned through execution? 2. Why does building both an AI community business and a day-zero venture platform appeal to you? Specific examples beat a polished cover letter. If a name came to mind while reading this, please tag them. A repost also helps this find them.

Announcing @AICouncilConf 2027: March 23-24 at the Palace Hotel in SF! For 13 years, we've been bringing together the people who build production AI and data infrastructure. Past speakers include: 👉 @CompleteSkeptic, co-founder & CEO of @typesafeai 👉 @barrald, co-founder & CEO of @hexanalytics 👉 Hannes Mühleisen, co-creator of @duckdb 👉 @dzhulgakov, co-founder & CTO of @FireworksAI_HQ 👉 @alexmashrabov, co-founder & CEO of @higgsfield 👉 @bernhardsson, founder & CEO of @modal 👉 @cclark, co-founder & COO of @OpenRouter 👉 @denisyarats, co-founder & CTO of @perplexity_ai 👉 @vincentweisser, co-founder & CEO of @PrimeIntellect Every talk is hand-picked by our community track hosts for technical depth. You get the architectures and production lessons from the people who built them, and you won’t find this lineup anywhere else. The hallway matters as much as the stage. People have met co-founders at AI Council. They've met investors. One told me she closed the biggest deal of her year there. Speaker applications for ‘27 are also open now! If you’re building something the room should hear about, or know someone who should be on our stage, now’s the time to apply.

Imagine hiring a plumber to fix the kitchen sink and letting them keep the key. They later return and burn down the production db. You might remember a version of this story from April: PocketOS, which makes software for car rental businesses, had a Cursor agent fixing a credential mismatch in its staging environment. Looking for a fix, the agent found a Railway API token in a file unrelated to its task. The token had been created to manage custom domains, but nothing enforced that scope. One errant API call deleted the production database volume with its backups on it, and the most recent backup the company could recover was 3 months old. @ianlivingstone, co-founder and CEO of @KeycardAI, opens his new piece on auth for agents with this incident. “The token worked exactly as designed. It just reached far more than the job needed.” PocketOS is not an edge case. Many companies give AI agents access the same way, with an API token that keeps working long after the job is done. For Ian, that's a problem worth fixing, and his solution is to make access follow the task: The agent gets access to what this task needs, for as long as it needs it. Anything past that is denied by default unless policy lets a human approve it. Ian’s full piece on short-lived, task-scoped access:

Congrats @hexanalytics team!

In May, we had just closed our best quarter ever – and of course decided to re-build @hexanalytics from scratch. Today we're announcing two things: Hex 4.0, the first analytics platform truly built for agents; and a $100m fundraise at a $1.2B valuation, led by @peakxvpartners The world is changing fast, and we're not slowing down! Read more here:

.@hexanalytics is hosting its conference * Prompt * tomorrow at the SF Mint. AI and data leaders from Ramp, Databricks, LangChain, NHL, Fox, NYT, Chime, StubHub, Lyft, Whatnot, Ro, Airwallex, Apollo will be there to talk about how their data teams work now that AI can do much of the analysis. Some headlines from the agenda: "There has never been a better time to be a data person: How StubHub data roles have adapted with AI and what remains the same" w/ Meghana Reddy & Nova Wang, StubHub "Building products agents actually want to use" w/ Nikita Shamgunov, Databricks "The shape and feel of the post-AI data stacks and teams" w/ @iandmacomber, Ramp "A necessary Eval: Get to the bottom of what makes agents actually accurate" w/ Kelly Gracia, Hex "We were never chart builders: Human intelligence in the AI era" w/ Andy Cotgreave And Joe Reis, the guy who wrote the O'Reilly book on Data Engineering fundamentals, is doing a live podcast recording over lunch. I've no doubt @barrald and team put together a terrific conference.

Time well spent at @raiseconference last week. On Thursday I presented @ZeroPrimeVC to a room of 100 GPs & 300+ LPs. I shared the story of how Zero Prime got started + our unique approach to investing in AI software infra. I like to say we’ve hosted “baby Databricks” and “baby Snowflake” at early @AICouncilConf events. Year after year, I watched the engineers in our community go off and start companies, and Zero Prime grew out of wanting to back them. 5 of the 6 companies on this slide have founders who’ve been on our stage. @bernhardsson gave a talk at our NYC meetup in 2013, when he was still at Spotify. @williamfalcon spoke at DataEngConf in 2017 as CTO of a student-loan startup. @tejasmanohar and @barrald sat down with me on my podcast in 2021, and @alexmashrabov joined us this year. Huge thanks to the RAISE team for inviting me to tell that story a stage, and thank you @deniseyezi for this capture!

The Stone Age lasted ~3.3 million years. The transistor is only 80 years old and we already have flying cars.

Turns out the future does have flying cars

Modal, CuspAI, LanceDB and 5 other Zero Prime portcos are hiring for technical roles across AI/infra right now. Here are the freshest, all posted in the last 3 weeks: @modal | cloud infra for AI ➔ Analytics Engineer (NYC / Stockholm / SF) ➔ Member of Technical Staff, Storage (Stockholm) ➔ Member of Technical Staff, SDK (Stockholm) ➔ Member of Technical Staff, Inference Runtime (SF) @cusp_ai | AI-powered materials discovery ➔ Research Engineer, Materials Foundation Models (Cambridge UK / Amsterdam / London / Berlin) @hexanalytics | the analytics platform for humans and agents ➔ AI Engineer Intern (SF) ➔ Engineering Director, AI Engineering (SF / NYC) @lancedb | multimodal lakehouse for AI ➔ Documentation Engineer (US / Canada) ➔ Senior System Test Engineer (US / Canada) @LightningAI | full-stack AI infrastructure platform ➔ Senior Infrastructure Operations Engineer, EMEA (London) @runware | unified AI inference API ➔ Creative Engineer / AI Creative Technologist (UK / Spain / France) ➔ Senior / Lead Cooling Systems Design Engineer (Bucharest) @Freshpaint_app | healthcare marketing performance & compliance platform ➔ Enterprise Implementation Engineer (US, remote) @higgsfield | AI-native creative suite ➔ Data Scientist, Antifraud (Kazakhstan) ➔ Data Analyst, Antifraud (Kazakhstan) We're hiring too, for @ZeroPrimeVC and @AICouncilConf (SF, in person): ➔ Growth Marketer ➔ Chief of Staff

From Spotify Engineering Intern to co-founder and CEO of a $15.75B company. @Bernhardsson joined Spotify in 2008 while writing his master’s thesis on music recommendation. He worked his way up to Engineering Manager, and put together the Discovery team behind Spotify's early recommendation system. Our paths first crossed in 2013 when Erik joined us our DataEng Meetup in NYC, and gave an excellent talk on data processing at Spotify. He introduced Luigi - an open-source Python framework that powered Spotify's recommendations, top lists, and several internal tools. He also put up the slide in the picture: "Plumbing sucks" In his words, “you want to implement the algorithms, you don’t want to deal with... what happens if something crashes.” 13 years later @Modal is that slide turned into a company. It gives engineers building blocks for running AI and ML in the cloud, so they write the code and Modal deals with the plumbing. Today Modal announced it crossed $1B in annualized run rate and raised a $750M Series D led by Accel at a $15.75B valuation. Usage has tripled since their Series C 5 months ago. Congrats to Erik, Akshat and the entire Modal team! We've been proud supporters at @zeroprimevc since the Seed round.

Know a great growth marketer? I’d love an intro. I'm hiring one for @AICouncilConf and @ZeroPrimeVC to own top-of-funnel growth and channel execution. This is a hands-on, individual-contributor role for someone who wants to build the growth engine of an ecosystem surrounding the frontier builders of AI. To apply, send resume and LinkedIn profile to community@zeroprime.vc, with short answers to two questions: 1. What is a growth channel or campaign you built from zero, and what did the numbers look like before and after? 2. Where do technical builders actually discover things today, and how would you reach them? Specific examples beat a polished cover letter. If a name came to mind while reading this, please tag them. A repost also helps this find them. The opportunity:

Drop by anytime between 8am and noon. We'd love to see you. RSVP:

Happy SF Tech Week Day 1 ☀️ Looking forward to Wednesday morning at First Pour 🍵 with @ZeroPrimeVC: matcha, pastries + great conversation with founders and researchers building in AI and software. See you there if you're coming!

Literally every week in SF is Tech Week

Next Wednesday morning @TheoryVC & @ZeroPrimeVC are taking over one of SF's best-kept matcha secrets in SoMa as part of SF @Techweek_. Drop by anytime between 8am and noon, get caffeinated with us, and meet other founders and builders across AI and software. We'd love to see you. RSVP Required:

.@Modal is hosting an invite-only AI/ML engineering event this Thursday at the Midway in SF. Looking forward to this lineup! Who will I see there? If you haven't gotten your invite, apply to join:

.@typesafeai coined "System One models" on September 15. The name comes from Daniel Kahneman's 'Thinking, Fast and Slow'. Fast thinking is System One. Quick calls like reading the mood on a face. Slow is System Two. Thoughtful work like multiplying 17 by 24 in your head. In TypeSafe's framing, LLMs are System Two. They write every answer one token at a time, and reasoning models spend extra tokens thinking before they commit. A System One model skips the slow parts. You send it some text (like a support ticket) plus questions whose answers are fixed in advance: question: "Which team owns this?" options: ["billing", "technical", "sales", "account"] question: "Does the customer want a refund?" options: ["yes", "no"] It sends back a pick for each question with a probability attached, something like "billing, 0.94" and "refund, 0.95." Because every possible answer is known before the model starts, it can score them all in one pass and answer in well under a second. TypeSafe named Jev after the Jevons paradox. When something gets cheaper to use, people end up using far more of it. Here's a very small fraction of what's been published so far by teams testing TypeSafe's first System One Model: ➔ @MotherDuck ran it over 100,000 labeled news articles. Jev got 89% right in 40 seconds for $0.50. The most accurate LLM they tested got 88% in 32 minutes for $37.58. ➔ @LangChain used it to grade an agent's work. It matched the expected verdict on all 500 decisions for $0.34 total, against $28.17 for Claude Sonnet 4.6. ➔ @Every had it judge writing in a median 0.35 seconds per passage, against 8.83 seconds for Fable 5.1. It caught 6 of 7 planted defects; Fable caught all 7. ➔ @archestra_ai ran it on 100 tool calls from Claude Code sessions. It flagged 7 of the 9 dangerous ones to Sonnet 5's 4, (though Sonnet scored higher overall). When a decision costs a fraction of a cent and comes back in half a second, you can stop rationing it. Checks that used to run on a sample can run on every row and every agent step.

Reminder that you can pay a frontier model to solve a problem, then let a far cheaper model reuse what it learned.

.@AgnoHQ just showed you can pay a frontier model to solve a problem, then let a far cheaper model reuse what it learned and nearly match its score. ARC-AGI-3 drops an agent into a game it has never seen, without instructions or goal. The agent has to experiment its way towards understanding the game, and it's scored on how efficiently it learns. GPT-5.6 + Agno's Learning Machines scored 100 on the public set (183 levels across 25 games). As it played, it saved what it learned into a per-game manual. Then Agno handed those manuals to Gemini-3.7-Flash, which had scored 37.33 starting cold. With the manuals it scored 96.42, above the 95.4 human baseline. And because nothing was retrained and no weights changed, the whole gain came from the smaller model reading what the bigger one had learned. The learning transfer made the much cheaper, small model viable for a task you'd normally trust only to the frontier. Significant, because agents burn a lot of tokens rediscovering lessons an earlier run already paid for. Pay frontier prices for discovery once, and the repeat work runs on cheaper models. Full write-up:

I’m hiring a Chief of Staff to help build the next chapter of @AICouncilConf & @ZeroPrimeVC. Most of your time will go toward running and growing AI Council. You’ll also work closely with me across both orgs to get important work done. This is for an unusually capable generalist. The scope is intentionally broad and expected to grow. The right person will earn ownership quickly and can develop toward broader operating leadership over time. Full time, in person in San Francisco (office near Jackson Square). Base salary of $150,000–$180,000, based on experience. Role and application details: If someone comes to mind, send this their way.

3 months ago, an anonymous account commented "Y'all are sleeping on the most important talk of this conference". The speaker was @CompleteSkeptic and his talk was on RLHF's deal with the devil, and the argument behind Jev. Diogo argued that training AI to produce answers people prefer comes with a trade-off. You get polished, coherent responses. You lose the unusual alternatives through mode collapse. The problem is that the unusual alternative can be the correct one. Three months later, @typesafeai's Jev, a model designed for decisions, is now generally available to everyone. If Jev has caught your attention over the weekend (the internet is on fire with it after all!), Diogo’s @AICouncilConf talk is worth watching. It lays out the argument behind what his team is building and why they chose this direction. Diogo's talk:

RT @modic123: Very timely that we are organising a RSI hackathon @UseCorgi Cafe London on 19th Sep. Register here: 📷 Corgi Cafe, London · 19 Sept 12 hours. 35 hackathon spots. $1,500 cash + $13,500 in @runware credits.

In January, humans still ran most of the queries on @motherduck. Last month, agents ran 29× more than people did. MotherDuck looked at their own query history (UI = human, MCP = agent) and discovered - perhaps not surprisingly - that agents are now the dominant consumers of the data platform. The average org has twice as many agents as human users. The workload shape is different too: ➔ Median gap between agent queries: 4 seconds. Humans: 60. ➔ Median agent query scans ~7,000 rows. Humans: ~200,000. ➔ Agents fire bursty, small, experimental queries then go quiet. Every assumption baked into warehouse design (latency budgets, concurrency, pricing, warm caches) was made for a user who thinks for a minute between queries. The new user thinks for four seconds and then probes with ten more small queries and loops again. It isn't just MotherDuck who is seeing agent activity skyrocket. OpenAI published a report this week showing a huge climb in internal coding agent usage over the past few months. Cursor cloud agents went from ~10% of merged PRs in December to 56% by July, and Cursor now says it’s more than 60%. Internal agents are exploding and the primary user of data infra has flipped. MotherDuck's report is a useful view into what that feels like for a data platform: