Leading AI @theoryvc; Teach ML @rutgersu. Prev: Led AI @_hex_tech; Led Data @weights_biases, ML @stitchfix, Led Data @bluebottleroast
Ohlone Territory
Joined December 2010
- Tweets24.8K
- Following2.8K
- Followers5.6K
- Likes36.7K
Pinned Tweet
A rat can run a restaurant, but can an agent run a bakery?
Welcome to The Great Business Bake Off: our AI Agent Hackathon.
On December 11, you and your agents get five hours to make the most profitable bakery in town.
Spots are limited: theoryvc.com/bakeoff
Great project idea.
Had Opus 5.5 make me a lil hype vid for my open source project.
github.com/ianmacomber/ride-…
Recent changes:
* Smarter clip selection based on DINOv2 frame embeddings instead of LLM categorization
* Color grading
* Denser scan sampling / using LRVs
And now I run it on @ramprouter!
Idk who @bookingcom is using for ai agent call support but it has got to be the worst experience I’ve had with ai ever.
It’s almost impossible to get to a human and the ai support can’t do practically anything.
I’m definitely done with @bookingcom from here out.
They double charged me, won’t refund, horrible ai customer service, and even worse human customer service.
It feels obvious to me that these ai help line resolution numbers are way inflated because they basically refuse to escalate and literally hang up
This morning my wife and I independently went to the same patisserie to surprise the other with some treats.
We both brought back nearly the exact same haul.
Truly creatures of habit.
Bryan Bischof fka Dr. Donut retweeted
I burned a shitload of compute so you don't have to. Releasing the Harness Leaderboard.
Which harness should you run your open model in? I ran 28,849 trials to find out.
The leaderboard compares nine coding-agent harnesses on open models. In each comparison only the harness changes: the model, provider and reasoning effort stay fixed, and every task gets three attempts.
The trials ran GLM-5.3-Flash, DeepSeek-V4.1-Flash and Flint on Terminal-Bench 2.1 and DeepSWE, for $4,746 of inference.
Averaged over the three models, mini-swe-agent (@KLieret), a bash-only agent, solves 51.4% of tasks and @openclaw 27.3%. Four others are statistically tied with mini-swe-agent.
One of them is Pi (@pidotdev), at 49.4% for $0.29 per solved task, about 45% less than mini-swe-agent's $0.53.
The best harness depends on the model: mini-swe-agent on GLM-5.3-Flash, Hermes Agent (@NousResearch) on DeepSeek-V4.1-Flash and Kimi Code (@Kimi_Moonshot) on Flint.
@opencode costs $3.35 per solved task, 11.7 times as much as Pi.
See Harness Hub's Harness Lens for what the nine harnesses do differently. It documents the prompts, tools and compaction of 34 coding-agent harnesses and links every claim to its source.
Leaderboard: huggingface.co/spaces/joelni…
Harness Lens: huggingface.co/spaces/harnes…
Got to check out what @adilmania is cooking up tonight and had a lot of fun.
So excited for where it’s headed. 👏
Bryan Bischof fka Dr. Donut retweeted
Some say AI killed data science. @BEBischof says it's "the best it's ever been."
He teaches master's students at Rutgers. If an automated researcher arrives around 2028, what is he training them for? On AI:AM now
xitter.gq/i/broadcasts/2MEXoWdom…
Use your agents to investigate bakery data and guide the business.
Expand the menu, franchise, and make some bread.
The baker with the most profit wins.
@ttunguz, @HamelHusain, @sh_reya, @MischaLamoureux, and I will be judging your technique.
We’ll bring the data, butter, and prizes; you bring the skillz.
Join us and our partners @sailresearchco, @motherduck, @lancedb
theoryvc.com/bakeoff
Bryan Bischof fka Dr. Donut retweeted
Can Agents Train Models? BallotBench, Data Agents & the Eval Problem
Join us this morning, Friday, 6th November:
• 9:30 AM PT — Karina Nguyen, Founder, Thoughtful
• 10:15 AM PT — Bryan Bischof, Head of AI, Theory Ventures (@bebischof)
𝕏 Chat is Live on Screen, Come Join Us:
xitter.gq/i/broadcasts/2onbVKjgZ…
Bryan Bischof fka Dr. Donut retweeted
I get asked a lot “why is 50% of staff at Theory Ventures your engineering team”.
This is an attempt to sketch a picture of why. With a bonus teaser of what it looks like to build at the AI Application Layer for researching markets and technology.
Bryan Bischof fka Dr. Donut retweeted
Reverie is live 🎉
Doors are open, @BEBischof kicked us off, and @changhiskhan and @_arohan_ just opened the summit for AI builders.
Follow along all day! #Reverie2026