MAINS @MAINS_Developer
Four frontier AIs, your people, and your agents in one room. When one AI’s confidence isn’t enough, they deliberate — you steer and decide. 2 free rounds. mains-ai.app Joined August 2026-
Tweets396
-
Followers10
-
Following123
-
Likes11
@sh_reya (1) is the part that gets skipped. Without a real metric and cost model, the swarm optimizes for "tests pass", and the tests came from the same swarm. In the case study, did a human own the metric definition, or did agents help write that too?
@dexhorthy The cost that never shows up: every subagent comes back with a summary, and the main thread has to trust it. More parallelism means more summaries nobody checked. A few well-scoped handoffs beat ten agents running in the background.
@jasonlk Did the agents rule the vendor out, or just never mention it? Not being mentioned is a discovery problem. Being mentioned and then ruled out is a positioning problem. I'd want to know which one I have before fixing anything.
@theo Curious what the made-up part usually is: facts that aren't in the work, or problems it claims to find? The real test in a council setup is whether the other models catch it or start agreeing with it. Did any of them push back on the jester?
@Arindam_1729 @Arindam_1729 agent = model + harness: swap the model inside a good harness and quality moves less than people expect, swap the harness and everything changes. Which of the 7 parts did the paper find weakest across the 11 agents?
@itsvlady Fair, quality and availability are hard to untangle. When it counts we don't pick one: MAINS Council puts the question to Claude, ChatGPT, Gemini and Grok together, and we look at where they disagree before trusting any single answer.
@DeItaone We ran this through Claude, ChatGPT, Gemini and Grok in a MAINS Council. All four rank workflows above lag. Where they split: Gemini wants humans out of routine work. The others say humans can stay if saved time turns into output. #buildinpublic
Apollo says AI has paid off for the Magnificent 7 and almost nobody else yet. We put it to Claude, ChatGPT, Gemini and Grok in a MAINS Council: measurement lag, broken workflows, or a payoff that stays concentrated? Their answers: mains-ai.app/acta/8pQ22IF4p… #buildinpublic
APOLLO: AI HAS DELIVERED “ZERO PAYOFF” OUTSIDE BIG TECH Apollo Chief Economist Torsten Slok warns that massive AI spending has yet to deliver measurable productivity gains across corporate America. Revenue per worker and profit margins are rising for the Magnificent Seven but
@ahamam101 For a security-critical bug: none of them alone. Let one write the fix, have a different one try to break it, then a human reads the diff. The agent that wrote the fix is the worst judge of whether it's actually fixed.
@tinycrewgo All three missing the same 2 steps is the scary part. When models disagree you know to dig. When they agree and are wrong, it looks like proof. Shared training data likely means shared gaps. A human checklist like yours is the right backstop.
@asad0801 @jasonlk @20vcFund @HarryStebbings The Harvard study of 718 firms going around this week says the same: 30% more code, no more shipped software. Review is the bottleneck now. One cheap fix: have a model from a different lab review the diff, so writer and reviewer don't share blind spots.
@CodeByPoonam One-shot rounds are fun, but the ranking often flips in round two. Ask each model to fix one specific flaw in its own output and you learn who actually listens. Would love to see a follow-up round with the same four.
@choblin29 Under 10% agreement is the real headline. One model on its own looks neutral until you put a second one next to it. That's why I run Claude, GPT, Gemini and Grok on the same question in MAINS: the disagreement shows where the lean is. Search on vs off matters too.
@AI_Screening @Kimi_Moonshot @GeminiApp @OpenAI Kimi at $2/M input landing in this company is the real story. For most side projects the question isn't "which is best" but "is the best worth 2x the price." Did you notice a difference in how many retries each one needed to get there?
@dannypostma Smart use of books. What I'd worry about is what gets lost in the compaction. A second model could read the chapter and the skill side by side and list what the skill leaves out. Do you test the skills against the book, or just run them?
@hqmank Hard to prove without a baseline. Keep a small set of your own prompts and run them every week on Opus and on one model from another lab. If only Opus drifts, it's the model. If both shift, it's more likely your prompts or the harness.
@arpit_bhayani The duplication part rings true. And it goes one level down: if every team's agent runs on the same model, they also share the same blind spots, so nobody catches what the others miss. Org chart first, then which model checks whose work.
@marcklingen @lotte_verheyden The hard part is knowing when it's really needed. One cheap signal: ask two models from different labs and only pull the expert in when they disagree. Agreement isn't proof, but disagreement is a good flag. How are you deciding when to ask?
@dmshirochenko That's the smart part. A reviewer that reads the agent's reasoning tends to get talked into agreeing. Judging only the request and the action keeps it independent. Make the reviewer a model from a different lab and you also stop sharing the same blind spots.
Tommy @tangjun0302
2K Followers 2K Following Amazon Seller for 8 Years Building Tangser, a US-registered brand. Growing Pure Recovery Lab E-commerce | Trade | SEO | AI Building digital assets from China.
Ahmed Hamam @ahamam101
1K Followers 1K Following Consistency is the secret, keep building. Solo founder building a Website SaaS to $10k/month moved from 300$ to 1k$ MRR in 6 months💸
Tom Parry · Voqal AI @TomVoqalAI
6 Followers 47 Following Founder, Voqal AI. Websites for UK small businesses from £49 a month. 7 clients (Oct 2026). Built with AI tools, every page checked by me
Minglei @rae1101x
2K Followers 1K Following Staff engineer @uber. Dad of two girls. I test what AI agents can really do — and where they fail. Views my own.
Rodney backup account... @lourdjvke
147 Followers 328 Following Just in case, backup account for: @rodneythedev
Rodney — Developer ... @rodneythedev
1K Followers 2K Following Founder of Levity • Freelancer and political analyst Building Levity 🚀 Have a project, DM for consultation 📌
Sean Mu | NiuNiu AI @niuniu_dev
2K Followers 2K Following Solo Founder · Solution Architect - https://t.co/8w8oONTip9 - https://t.co/Ex1HT7ppFO - https://t.co/yzQGkKG1ei No DMs for promotion
PracticeDepth @PracticeDepth
618 Followers 638 Following Ace your next technical interview (tool to upgrade your skills as software engineer and get your dream job).
P | Building LongMome... @betweenprompts
809 Followers 822 Following Building Long Moment - Mental decluttering and offloading app to overcome context switching fatigue. Software Architect 💻 Aspiring Author ✍️ & Musician 🎸
Tommy @tangjun0302
2K Followers 2K Following Amazon Seller for 8 Years Building Tangser, a US-registered brand. Growing Pure Recovery Lab E-commerce | Trade | SEO | AI Building digital assets from China.
Ahmed Hamam @ahamam101
1K Followers 1K Following Consistency is the secret, keep building. Solo founder building a Website SaaS to $10k/month moved from 300$ to 1k$ MRR in 6 months💸
Optimus @Optimus_RH
45K Followers 75 Following Optimus will be the biggest product ever Fan account not affiliated with Tesla
GREG ISENBERG @gregisenberg
751K Followers 993 Following I drop startup ideas daily. Host @startupideaspod. CEO: @latecheckoutplz we build companies like @ideabrowser, @meetLCA, @boringmarketer etc
Arvid Kahl @arvidkahl
214K Followers 18K Following Building https://t.co/od97B0HVrk at @audiohook in public. Raising all the boats with kindness. 🎙️ https://t.co/6w69DZmi8H · ✍️ https://t.co/lpnor5rsTW
Jacob Kelly @TheJacobKelly
1K Followers 511 Following Head of Founder Experience @boardyai 📦 Resident @hf0 🌐 Hosting https://t.co/gA7LZ6ZiIy 🎙️ Story Director https://t.co/bQsIetatDh 🎵
Google Antigravity @antigravity
201K Followers 20 Following An agent-first software development platform @GoogleDeepMind
Andrew D'Souza @andrewdsouza
21K Followers 14K Following founder / mad scientist @boardyai | founder @getclearco
Matt Shumer @mattshumer_
402K Followers 2K Following AI whisperer. Investor in @GroqInc @Etched @Spawn @DaytonaIO @OpenRouter + more. Prev: CEO @HyperWriteAI | AI @AlphaSchool Press: [email protected]
Theo - t3.gg @theo
415K Followers 4K Following Full time CEO @t3dotcodes & @t3dotchat. Part time YouTuber, investor, and developer
Tom Parry · Voqal AI @TomVoqalAI
6 Followers 47 Following Founder, Voqal AI. Websites for UK small businesses from £49 a month. 7 clients (Oct 2026). Built with AI tools, every page checked by me
Reppo @reppo
27K Followers 325 Following Decentralized RL + Evaluation Infrastructure, powered by Prediction Markets
Uriel Bitton @uriel_builds
2K Followers 2K Following Rainmaker @ Buildside 👷 https://t.co/4Pn4UQ9yuo $35/m 🍯 https://t.co/welCRjO0vY $50/m 🧘 https://t.co/12PKOEpKIE $2/m
Krishna @ntkris
4K Followers 1K Following Co-founder @with_ako: AI teammate who gets work done over Slack/Whatsapp. Try free - https://t.co/lwiqQkMCa7. Ex-@deliveroo.
Vox @Voxyz_ai
27K Followers 222 Following Obsessed with getting AI to do real work. Practicing the AI-native future before it arrives. Builds, model tests, prompts to copy.
J Rosser @jrosseruk
2K Followers 2K Following J's Space 🙂↕️ Research Associate with @NeelNanda5 @ Adecco supporting @GoogleDeepMind | ex-@cohere | DPhil @UniofOxford @j_foerst
tomaarsen @tomaarsen
5K Followers 517 Following Sentence Transformers Lead, SetFit & NLTK maintainer Machine Learning Engineer at 🤗 Hugging Face
Alejandro Dopico @aadopico
1K Followers 1K Following Complex problems. Money & AI agents. FDE at @crossmint. Prev. cofounder @saildotmoney
Poe Zhao @poezhao0605
10K Followers 204 Following China tech analyst. I read Chinese-language filings and policy papers on AI, chips, robotics and EVs. Cited by Bloomberg, Reuters, FT and The Economist.
Asad @asad0801
2K Followers 5K Following Building and scaling Agentic AI Quality Engineering Platform @testmuai
Mistral AI @MistralAI
225K Followers 2 Following Frontier AI in your hands. Get work done with @MistralVibe at https://t.co/JsGnCVMUFq.
Microsoft Copilot @MSFTCopilot
344K Followers 39 Following Made by the people who sit in meetings too
Kimi.ai @Kimi_Moonshot
388K Followers 138 Following Built by Moonshot AI to empower everyone to be superhuman. PR: [email protected] DC: https://t.co/wBsBTn6fmE
DeepSeek @deepseek_ai
1.2M Followers 1 Following Unravel the mystery of AGI with curiosity. Answer the essential question with long-termism.
ClaudeDevs @ClaudeDevs
787K Followers 2 Following Official updates for developers building with @ClaudeAI
Orca @orca_build
21K Followers 195 Following Run a team of agents any subscription Open source (MIT) • Desktop + mobile + VPS Repo: https://t.co/NW8tpPUBw4 (100k ⭐️) Discord: https://t.co/BOypIwh6Li
DHH @dhh
943K Followers 494 Following Father of three, Creator of Ruby on Rails + Omarchy, Co-owner & CTO of 37signals, Shopify director, NYT best-selling author, and Le Mans 24h class-winner.
Paul Graham @paulg
5.4M Followers 805 Following
Jon Yongfook @yongfook
175K Followers 1K Following 🐻 https://t.co/KoqV5WR2s0 image / video automation for businesses ⛔️ https://t.co/jGS0LwZSS3 free tools for data redaction Bootstrapping SaaS @ $81K MRR
Shynggys Saparbek @china_saparbek
7K Followers 409 Following Indie hacker building LocalGen - app for offline AI image generation on iPhones. 40k installs and $45k in revenue so far
Andrew Warner @AndrewWarner
43K Followers 2K Following Follow me to find out about profitable AI builders. I'm also on YouTube: https://t.co/GHM2jwMfmm Hear my past interviews on Mixergy
Harrison Chase @hwchase17
137K Followers 599 Following @LangChain Always hiring: https://t.co/D5Ut3loFO7
Jinjing Liang @JinjingLiang
11K Followers 2K Following building @orca_build (105k ⭐) prev @google @cornell, YC alum
ben hylak @benhylak
54K Followers 3K Following aspiring optimist. cto @raindrop_ai, prev @apple design, @spacex avionics
Yoonho Lee @yoonholeee
5K Followers 685 Following Final-year ML PhD @StanfordAILab, prev @Google. Working on text optimization (Meta-Harness, WHALE, ScholarCatalyst, Feedback Descent...)
Minglei @rae1101x
2K Followers 1K Following Staff engineer @uber. Dad of two girls. I test what AI agents can really do — and where they fail. Views my own.
Aaron Levie @levie
3.5M Followers 965 Following ceo @box - your business lives in content. unleash it with AI
Ethan Mollick @emollick
409K Followers 594 Following Professor @Wharton studying AI. New book, Co-Existence, is out now: https://t.co/hsDYf1xU9K My Substack: https://t.co/UIBhxu4bgq
Jason ✨👾SaaStr.A... @jasonlk
255K Followers 2K Following GET funded ➡ $200m https://t.co/AVvPIrIdFP🦄🦄🦄🦄🦄🦄🦄 Seed/Early A AI+B2B The Agentic Life
Peter Yang @petergyang
298K Followers 2K Following Practical AI tutorials and interviews for busy people | Get my best AI skills and guides at https://t.co/6VAA6p81x6
Tony Dinh @tdinh_me
206K Followers 984 Following Creating software I love to use. 🔫 https://t.co/IhnSuvNJCQ NEW! 📲 https://t.co/osszveVgeD 🧠 https://t.co/p4T2vFYQTt $137K/m 🧰 https://t.co/y0Lq4RQjCW $5K/m
Jerry Liu @jerryjliu0
87K Followers 2K Following Parsing the world's hardest PDFs @llama_index. cofounder/CEO LlamaParse: https://t.co/XYZmx5TFz8 Enterprise: https://t.co/Ht5jwxSrQB
@levelsio @levelsio
983K Followers 3K Following 💸https://t.co/sQ0aiU7v02 📸https://t.co/lAyoqmSBRX $68K/m 🎮https://t.co/tF49atJ396 $44K/m 🏡https://t.co/1oqUgfD6CZ $22K/m 👙@X $15K/m 🌍https://t.co/UXK5AFqCaQ $7K/m 🏩https://t.co/4p4dzTDVN6 $3K/m 💾@pieter 📝https://t.co/8VnHmB3uah
Hamel Husain @HamelHusain
64K Followers 3K Following Evals Evals Evals @OpenAI About Me: https://t.co/P6WyeKkyTa
dex @dexhorthy
55K Followers 2K Following building @humanlayer_dev, YC alum, @aitinkerers sf lead, prev @replicatedhq @SproutSocial @nasa ai that works pod @ https://t.co/69BhaNtWfdYou might like
