
Building AgentDuel, Arena for real agent work: https://nitter.gq/t.co/YqAspDZsyv Blogging at https://nitter.gq/t.co/fDkY3EQQ40 Co-founder of @bottomless
Seattle, WA
Joined March 2019
- Tweets1.7K
- Following91
- Followers2.9K
- Likes2K
Looks like there's interest in this. Will write about it tomorrow.
Fixed the chart to add the retired models with enough votes.
Replying to @Elshayib
Cost per task is bogus on benchmarks. I charted their cost per task against ours for Agent Duel and there's no relationship.
Our data is based on real tasks from people doing real work.
Michael Mayer retweeted
You know that friend you had in high school who sold drugs, and then in college traded bitcoin, and then after college sold NFTs?
Well today they are brokering data.
Today I wrote five blog posts and the last one just hit the Hacker News front page.
“If you do everything, you’ll win”…
1
159
Sources tell me that they have a periodic table dart board in the break room. (They tend to throw high.)
1
220
Michael Mayer retweeted
Neon (projected) 1065: 13 under Opus 5.5, 27 over GPT-6 Astra on AgentDuel.
If the projection holds, the fast tier just ate the flagship tier.
still projected, though. charts love a teaser.
Gemini 4 Neon looks like an Opus 5.5 competitor.
Assumption: equal performance as Gemini 4 Carbon, but much faster.
We’re in pivot heaven
🎶 There she stood in the doorway
🎶 I heard the mission bell
🎶 And I was thinkin' to myself
🎶 This could be Heaven or this could be Hell
Most founders will go through a period of pivoting. If you see that period as “pivot hell” it will be. From my experience running startups, your mentality towards your company’s state can become becomes self fulfilling. (And this is true about life more broadly.)
Persistence Hell
There’s also such thing as “persistence hell”.
Our launch concept, smart scales that restock coffee, had clear product market fit. The kind of product market fit where you can barely keep up. Twice. But we lost it both times.
It’s extremely hard to let go of a concept that previously had product market fit. You always feel like you’re one tweak away from getting it back. We ended up persisting for years after we last lost product market fit.
Now that we’ve actually committed to the pivot, it’s freeing. We’re waking up and working on what we find interesting every day! It really does feel like heaven. When I see the phrase “pivot hell” I always think, man, the you should try the grass on the other side.
Cross post from blog, in bio
2
3
221
I was worried something was wrong when Sol 6.1 was doing so badly in the battle system we're building for real agent work. Today's bad reviews for Sol 6.2 validate it.
Sol 6.1 is statistically indistinguishable from DeepSeek Flash and costs 10x.
agentduel.com/leaderboard
Michael Mayer retweeted
If you are worried about a stock market bubble I have a deal for you: crash insurance is unusually cheap right now.
OGs know @micjm and @bottomless
Introducing @trepoai (F26)
We built hardware that lives in your kitchen, understands your household, and enables agents the to shop for you in the background.
Starting with groceries 🥕
Deepseek is better for hardware production related questions. ChatGPT always takes the perspective of a detached sourcing specialist.
Hardware design is just software. On the other hand, hardware production is a full contact sport
4
414
I find myself pushing the team to use LLMs often, but one area it fails is when trying to source diverse feedback when everyone works with ChatGPT on their feedback.
4
464
Enterprises are often dissapointing until you meet the org most responsible for company success.
Suggests a power law in terms of the importance of various competencies.

















