
Ender's Game for Autonomous Systems. @exialabsai. Visiting Fellow @HooverInst. Ex-Meta, Amazon, Riot Games. U.S. Army veteran 🇺🇸
Bellevue, WA
Joined October 2011
- Tweets747
- Following702
- Followers2.2K
- Likes17.5K
Pinned Tweet
We are excited to announce Blue, your Copilot for Military Operations. Commanders who decide faster, win faster. Blue automates labor-intensive, “science of war” decision making inputs, like terrain analysis, mission analysis, decision support templates, so commanders and staff have more time for the art of war and making better informed decisions.
War has become increasingly digital, decentralized, and fast – leading to compressed decision cycles. The future of military operations will demand innovative approaches to planning, rehearsal, and execution to outpace adversaries and seize the initiative.
Jonathan Pan retweeted
AI may well be able to write massive software projects “from scratch”, but that almost certainly isn’t the fastest / highest performance / highest security / most token efficient way to do things. Good building blocks matter.
In addition to RL’ing the agents on final task loss, we should try to get them to regularize the building blocks as well. What would the standard library look like that makes the agents most effective?
It is likely at the limit that it turns out to be some alien organization of code, but given the existing human-bias in training, I bet that in the near term it would be lovely code that old programmers would appreciate.
I suspect the same could be said for mathematics, but that is out of my lane.
Jonathan Pan retweeted
hey all - update for SR008:
It's now the FINAL weekend to apply for our next residency in San Francisco. Applications close this Sunday, Nov 1 at 11:59 PM PT.
Here's the link:
speedrun.a16z.com/apply
How it works:
- you apply
- we review, reach out to get introduced, etc
- if there's a fit, we handshake and sign
- we wire our investment right away!
- plus you get access to >$10M in credits right away
- you'll onboard + start working with us immediately, have access to working out of our SF offices, etc
The official residency starts Feb 2027 where we'll have speakers, office hours, and much more. I plan to be on site and look forward to meeting many of you
You might ask- who's the best fit for this? We want to talk to you as early as possible, and that means:
- no company yet
- no deck
- no huge insight tweet thread
- sometimes not even fully committed yet
You don't need the perfect idea. We can start working together when you're iterating and getting the idea off the ground. That's the whole point of speedrun - it's our way of backing you from Day 0.
More official stuff below:
- Deadline is this Sunday, Nov 1 at 11:59 PM PT (this weekend!)
- SR008 runs Feb 1 through April 18, 2027 in San Francisco -- twelve weeks of firesides, office hours, and sessions, closing out with Demo Day on April 13 in front of thousands of investors
- Capital: up to $1M with a two-week close target, plus $10M+ in credits from 200+ partners
- Customers: SR Insiders gets you in front of 10,000+ buyers and operators, including 150+ C-suite execs, plus a full day of live demos at AI Faire
- Talent: our Talent Network of 171,000+ engineers, designers, and product people. 230+ hires so far, $0 in fees. Plus visa, banking, and housing help if you're coming from abroad
- A team of ~40 battle-tested investors and operators who help cover your gaps without adding headcount -- GTM, marketing, brand, talent, fundraising, and more
- a16z speedrun is super selective -- fewer than 0.4% of applicants get in. You'll join a tight-knit community of founders from 20+ countries who stay involved as speakers, advisors, and often each other's first customers
- Still in school or still finding your idea/team? Check out the alpha fellowship too -- up to $250K for students, recent grads, and dropouts
this is lovely - used airbnb over 20 times this year, this was the number 1 annoying thing and I'm glad they're fixing it
nytimes.com/2026/10/22/trave…
I made the mistake of giving my dog some wagyu steak and he's been following me around for an hour non stop pawing me for more (I already gave him half an entire steak) 😭 – at Beaux Arts Village, WA
Every defense tech firm has been offered money from Chinese or Russian affiliated sources, including us. It would be nice to use that money to accelerate our efforts, but it defeats the purpose of America rearming and reindustrializing.
My issue isn't with what one founder did, but the culture that enabled him to take any money when convenient. btw, we would never have heard about this if there wasn't a dispute. Now it's "Russian/Epstein" money (his investor's words, not the founder).
Jonathan Pan retweeted
We'll be spending a lot more time trying to understand the outputs of language models. A few thoughts, tips & tricks:
Writing. Something I've had success with: Ask your LLM to explain something in ASD-STE100, it's a controlled language specification originally developed for aerospace maintenance documentation. LLMs well-versed in this language and it comes with heavy constraints on clean writing style that I often find a lot more readable. Sometimes I've tried to soften it a bit e.g. ask for "80% of the way to ASD-STE100" because the spec is quite stringent. But even better:
Diagrams / images. Instead of writing, ask your LLM to create a diagram. These can be a lot easier to process, parse, and understand. But even better:
Web pages. Ask for output "in HTML" to get a beautiful, interactive webpage. LLMs are getting really good at frontend and can create beautiful experiences, animations, etc. But even better:
Explainer videos. The output format I am most bullish on is fully custom / bespoke explainer videos generated on any arbitrary topic. Experiment with things like "Create a 3b1b style video explainer on X. Use my ElevenLabs API key for audio narration". (you'd need an API key for the latter or you can ask your LLM to find you decent free alternatives that use your local compute). This is actually starting to work!
In summary:
- As LLMs get better, they will do more and more of the legwork autonomously, and a lot more of our work will rise up the abstractions into oversight and understanding.
- Luckily, LLMs can help here too because as intelligence and code are increasingly abundant, you can ask for large, custom, discardable software artifacts (e.g. web apps, video explainers) that would have never made sense to create before. Push the boundaries here and you'll be surprised.
I turned 18 only two days before 9/11. Every year, I learn more about the tragic and brave stories of the victims and first responders. Last night, I was digging around my emails and hard drives to see if I could find anything new and I found an email I sent to another freshman in college about joining the Army. He just finished his active duty service and was now in the New York National Guard, here was his response:
"Army life, especially the Infantry, is a lot of hard work. As an officer expect long hours and little sleep. Aside from your decent pay don't expect much else. Physically you have to be able to do everything your men can do and then some. For them to trust you and follow your orders while under fire you must get dirty with them. The best leaders I've encountered always seemed to be men that cared more for their men then themselves."
It's surreal to dive into some of the thoughts of 18 and 20 year olds back then. Despite all the tragedy and sadness, our collective response to 9/11 makes me very proud to be an American.
Jonathan Pan retweeted
We've unlocked a product milestone in Blue: the ability to integrate drones directly into missions (simulation then orchestration).
✅ Exposed an interface that allows Army/DoW platforms to ingest our data, query our feeds, and task our assets programmatically.
✅ Identify NAIs based on environmental and operational data.
✅ Simulate drone COA before execution.
✅ Confirmed tasking of our asset to recon an NAI in California from a user in Europe.
Jonathan Pan retweeted
We're excited to continue our partnership with @LockheedMartin and work with old friends like 4C Strategies to deliver next-generation modeling and simulation technologies for @NATO.
Replying to @LockheedMartin
@LockheedMartin has advanced to the second sprint of the @NATO Next Generation Modelling and Simulation competition alongside industry partners @4C_Strategies, @ExiaLabsAI and RB Advisory. The ACES software suite is the foundation of the solution.
Exia Labs has joined partners like Microsoft and Google in signing the "Open Weights and American AI Leadership" letter.
Efficiency is everything for military and intelligence workflows, especially at the edge. Not every prompt requires the firepower of Mythos or GPT-5.5. While mission-critical targeting workflows might need the reasoning capabilities of a frontier model, simpler workflows like generating a route or querying a logistics report can easily be handled by open-weights models like Inkling or Kimi.
microsoft.com/en-us/corporat…
Every time I go to the park and I see hundreds of people looking at their phones (or the telltale double phone), it's always because they are playing Pokemon Go. Crazy how people are still loving ther game 10 years after launch – at Clyde Hill, WA
The ultimate disrespect in the workplace today? Outsourcing your thinking to AI and forcing your co-workers to review your slop. If you don't start with a coherent thought, no frontier model can cover that up for you. Who else hates this?!?!
Defense Founder Note 4 of 20: The Defense Acquisition Timeline
Right now, there is a mad scramble within the government to spend end-of-year (EOY) money, and a significant portion of those funds is flowing through defense consortia.
Let’s zoom out for a second. Despite the endless podcasts, memos, and webinars proclaiming that defense acquisition is speeding up, it remains notoriously slow—especially for early-stage startups.
Achieving product-market fit in defense requires three things: alignment with the end-user, alignment with the buyer, and alignment with timing. Timing is usually the trickiest variable. As a startup, you spend most of your time in a direct feedback loop with the end-user to drive product development. If you’re wise, you also keep the buyer informed and aligned. But just like in commercial enterprises, military personnel and teams change frequently. For example, my team was once down-selected for a contract, only to be met with months of silence. We eventually found out through the press that the specific acquisition office had been reorganized, leaving us in limbo.
Returning to the current calendar: July is generally the final month to capture EOY funds, submit proposals, and shape opportunities. By August, defense acquisition quiets down significantly as Congress goes on recess. And since government shutdowns have become a standard political tactic, September always carries the looming threat of a government standstill.
This compressed timeline is exactly why many defense organizations are currently funneling money through consortia (Topic 5 of 20). Defense consortia are organized around specific technological domains, such as ground autonomy or modeling and simulation. The government issues a requirement to the consortium, members submit proposals, and the selected companies are awarded contracts via Other Transaction Authorities (OTAs).
While consortia and OTAs were designed for speed, they largely operate in the dark. This lack of transparency has become so pronounced that the U.S. Senate recently passed the Stop Secret Spending Act, which would require agencies to disclose OTAs just like standard federal grants.
For early-stage startups without a platoon of capture managers, this opacity is a major hurdle. It is incredibly difficult to gauge whether a consortium solicitation is a genuine open competition or just a contracting vehicle for a pre-chosen candidate. You can spend dozens of hours on a proposal, only to be completely ghosted; consortia aren't even required to notify you if your bid wasn't selected.
It gets even crazier. One consortium we worked with had a dashboard showing proposal statuses. When we asked for an update on a February submission, their response was: “We’re removing the feature that shows you the status.”
Replying to @SenJoniErnst
😨The Pentagon spent $77.5 BILLION through "Other Transaction Agreements" over the past five years.
But this spending isn’t public on USASpending.gov
My included Stop Secret Spending Act forces transparency so taxpayers can see exactly where their money is going.















