MEGA

X research — MEGA instructors

Original markdownProgram overview

Pulled 2026-10-19 ~9:40 PM ET via the bot's X read tools (search_posts_all / get_users_posts / get_posts_by_ids). Posts quoted verbatim (truncated at X's note-tweet limit where noted). No X activity for John Lindquist matched topic search; his timeline was read directly.

Theo Browne (@theo, id 786375418685165568)

  • 2026-10-19 — 2112331734272593922 — Can I be real for a sec? I'm super nervous about this launch. It's something I never thought I would do, and I want to be as transparent as possible. This course is $1,300. You should not buy it. Let me explain. I've done YouTube for over 5 years now. I've never done a paid https://t.co/jJ0Y9r9Wgu
  • 2026-10-19 — 2112310904532713628 — If you enjoyed this and want something deeper and more focused on "how to use AI at work", the course I'm building with Mega starts on the 26th and I'm really proud of it. Check it out at https://t.co/zU6GmpLZer More info: https://t.co/TtrcL0c5rz
  • 2026-10-19 — 2112309129545150755 — @rohit3a If you've seen all my videos, there won't be too much "new" here. It's way better organized and much more refined. My war on memory systems + actual deep dives into "how I make my agents do what I want them too" is a bit novel, but you can piece together most of it from my vids
  • 2026-10-19 — 2112307060910227942 — Really excited about the courses I recorded for Mega. I love YouTube to death, but lessons like these would never perform there. Excited to reveal a lot more about how I think and build :) ask your boss if you can join us! https://t.co/kAd8svCoPn
  • 2026-10-19 — 2112089045979615338 — @marcos_placona @t3dotcodes We don’t do anything special here. Whatever your harness pulls in, you get :)
  • 2026-10-16 — 2111245697815302303 — @thsottiaux > "We achieved this by reusing context across subagents, keeping context cached longer, and improved request routing that allows us to increase cache hits." Transparency here is dope, thank you! Exciting changes 🙏
  • 2026-10-16 — 2110937874384511156 — Normally I would try to help them fix things. I’ve tried for almost a decade. The world is moving faster, and they’re moving slower. It’s one of the biggest bottlenecks in my life. I legitimately think I can build something 10x better in a few weeks. Will spin up some agents on
  • 2026-10-15 — 2110713238593908868 — @Barthazian In the 24 days since Opus dropped, I shipped... - I built ts-rust, a complete rewrite of the Typescript Compiler in Rust that is 2-3x faster than the official version - Got 245 prs into T3 Code, an AI coding app with over half a million users. I shipped over 60 PRs in the last https://t.co/lvqnMxayID
  • 2026-10-15 — 2110706613493661800 — @appfactory @amorriscode Already kind of have this with the new "working" section. I want to go all the way and give agents a tool to "go back in the inbox", been starting to experiment with this
  • 2026-10-15 — 2110624230169776466 — @The_Contextian @thsottiaux > Shouldn't the harness already be aware of what model it's using? The model doesn't know what model it will be when it's trained. Nothing to do with hallucination. Most harnesses just tell it through the system prompt. Codex doesn't. It's stupid that Codex doesn't do it.
  • 2026-10-14 — 2110189831804604640 — @oglassdev File a PR and I'll tell my agents to figure it out
  • 2026-10-13 — 2110136892738195928 — @AniC_dev macOS is terrible for running agents Mac’s are the only decent computers to daily drive on your lap
  • 2026-10-13 — 2110106055984529564 — @mweinbach A shame they have the worst harness and also the worst record for banning people for using subs in other harnesses 🙃🙃
  • 2026-10-12 — 2109553016127045671 — @m_chirculescu @buildwithsid Omarchy is meant to be looked at. Ubuntu is meant to be ignored. Using omarchy for a headless agent box is a failure to understand omarchy, agents, and headless boxes.
  • 2026-10-12 — 2109544074705621463 — I may cave for a "math tool" in the future that allows agents to specifically call latex syntax for parts of a reply. I will NEVER ruin our markdown rendering with this fucking garbage because the math world is special and insists their cursed fucking syntax get normalized.
  • 2026-10-12 — 2109509406127956224 — @mweinbach @thesherlocker 🙃 If you can make my agents approve it, we'll see what we can do.
  • 2026-10-12 — 2109481941112430831 — @uwaooo @Josikinz I'm sure you enjoy nix a lot! I have things I love too, personally I'm a huge fan of CachyOS, zsh, zfs and lots of other things that agents don't benefit from. Agents aren't gonna figure out your nix setup. Keep it on your personal machine, give agents what they know.

Angie Jones (@techgirl1908, id 2932738834)

  • 2026-10-17 — 2111581244035133725 — I hear you all cheering for model improvements, but can we take a moment to appreciate just how good these agent harnesses are getting? ChatGPT Work is goated!
  • 2026-10-15 — 2110802970095071560 — YOU GET AN AGENT! YOU GET AN AGENT! EVERYBODY GETS… themselves to San Jose 😂 Can't wait to see you all in person! There may even be a lil something under your seat... just kidding lol but I do have a 25% discount for you ❤️ https://t.co/iAbm5OZIKN https://t.co/URIEw7ansC
  • 2026-10-14 — 2110416594409062510 — Asked my agent for next steps... apparently one of them was socializing with humans https://t.co/mJ7OHjwxqO
  • 2026-10-13 — 2110066913871286456 — I trust my agent with a lot. But I don't trust it to spot the homie in the hallway track and yell "HEYYYY!" with the appropriate level of enthusiasm 😂 https://t.co/h6P8gqUePi
  • 2026-10-10 — 2108989477544722640 — I've been on the road since August, and just came home to record my MEGA Dev course on Building Agents. I'm so excited about this new course! This is not one of those show-offy courses about making little personal agents for yourself. You don't necessarily need a course for https://t.co/d7msF06De2
  • 2026-10-10 — 2108905809295937858 — if you're gonna demo an agent booking a flight, this is a much more compelling use case https://t.co/MdYyjRGIZH
  • 2026-10-06 — 2107492870835880263 — The tester in me lovesss this 😍 New project, Era by @Eon_io_ lets you test your agents across apps like Salesforce, Slack, and Jira with connected data across those systems. A simulated company, accessible through MCP. Brilliant! https://t.co/46XSUaTPnF
  • 2026-10-06 — 2107492452294701144 — The tester in me lovesss this 😍 New open source project, Era by @Eon_io_ lets you test your agents across apps like Salesforce, Slack, and Jira with connected data across those systems. A simulated company, accessible through MCP. Brilliant! https://t.co/46XSUaTPnF
  • 2026-10-03 — 2106413044259545421 — Preparing the program for MCP Dev Summit Nairobi and these talk submissions are WOW! Pretty much every single one is focused on some critical real world problem not just "wouldnt it be cool if an agent did this". This is going to be an amazing conference! Looking forward to https://t.co/Rz9ZbOjCdJ
  • 2026-10-02 — 2106045938821112161 — If you haven't explored decision models yet, here's your chance. @johnlindquist's livestream will show Jev in action. This is a great intro before our Mega Dev course where I'll teach you when and where to use decision models when building agents. https://t.co/LmNj4jzdR3
  • 2026-10-01 — 2105654760296558720 — ask your agent about AGNTCon + MCPCon in San Jose. Super curious to know which sessions it thinks you should attend. https://t.co/yyLDUq1WI2
  • 2026-09-30 — 2105419471905304872 — dang, i was about to post to tell y'all to grab your tickets for MCP Dev Summit Toronto but the team just told me we're sold out!!! Can't wait to see you all and geek out on all things MCP! https://t.co/mKoUFFn8xq
  • 2026-09-30 — 2105402263158329553 — we got agent swag y'all! 🤖 https://t.co/eKSMQReuUY
  • 2026-09-29 — 2104973526826725703 — Meet Ola. One of the folks leading the Skills Over MCP working group. Ola and other volunteer engineers worked on extending MCP to discover and fetch Agent Skills Happy to see this accepted as an official extension 🎉 Ola gave us a break down of how this all works at AGNTCon +
  • 2026-09-29 — 2104970385783771493 — Meet Ola. One of the folks leading the Skills Over MCP working group. Ola and other volunteer engineers worked on extending MCP to discover and fetch Agent Skills Happy to see that this accepted as an official extension 🎉 Ola gave us a break down of how this all works at
  • 2026-09-23 — 2102791576460365972 — accidentally said automated tests instead of evals and they kicked me out of SF https://t.co/JOUt0ugZdm

Kent C. Dodds (@kentcdodds, id 389681470)

  • 2026-10-20 — 2112333064718426623 — So well said. Get your boss to buy you a ticket to Mega. We'll see you next week. https://t.co/7tpVlF86S7
  • 2026-10-19 — 2112330935031279870 — This is how you do product engineering now... https://t.co/pUtLwTUcS7
  • 2026-10-19 — 2112319939600376247 — @ByJavidShaikh come to mega and I'll show you!
  • 2026-10-19 — 2112046788395880454 — @pythonics00 Fair point! Verifyability is the key here and that would be a great place to invest some effort to close the agent loop and make it more autonomous 👍
  • 2026-10-19 — 2112046522804138376 — @YuriSulyma I've not written much library code since models got really good this year, but if I did I think I would still have the agent write all the code and I'd just ask it to show me code samples and we'd discuss the trade-offs of APIs in chat rather than hand-writing anything.
  • 2026-10-19 — 2112037485593117166 — I used to have a way in Kody for you to generate UI and display it inline in the chat (using the MCP Apps spec). It was cool, but I ripped it out because all the agents you want to use will implement a first-party way to do this. Kody is for things useful outside of chat. I may
  • 2026-10-19 — 2112035944526192973 — How much of the code you ship to production is written by an agent?
  • 2026-10-18 — 2111648471405252650 — @jessethanley @aarondfrancis @kettanaito @spamhaus Here's a summary of our findings and what we did about them from my agent. Thanks again Jesse! And thank you for the bat signal Aaron! Oh, and as a result of these findings, I'm going to do a purge of inactive subscribers on Kit (after asking them whether they want to stay) so
  • 2026-10-16 — 2111187810484887964 — Start with curiosity, not a solution. @ttorres on getting into the product conversation without picking a fight. https://t.co/fehyuwqyHA
  • 2026-10-16 — 2111177829463511468 — @marchoeffl As long as I'm using an agent that can run a CLI then that's fine, but maybe I'm not. OAuth is best for both of them.
  • 2026-10-16 — 2111119482244780241 — @KentonVarda I just reminded my kids yesterday that we are nice to AI chatbots not because they need us to be nice but because we need to be nice. My son said, "Yeah we need the practice." 🥹
  • 2026-10-16 — 2111111899005997505 — @belchugov @kodykoala this would be a fun idea to add but instead why don't we have you tell him how to set up his agent with Kody and then fork it in his own account so he can customize it however he likes?
  • 2026-10-15 — 2110835899596681368 — @ryanflorence Yep, mine do a good job of looping me in when it's time to make a product call or we're going to do a one-way door thing.
  • 2026-10-15 — 2110825482400969146 — @ClankerTees carries over and your agent can do migrations too
  • 2026-10-15 — 2110815577170378890 — I'm planning a video right now to talk about this exact thing. I'm calling it "invisible software." When it becomes so easy to generate a custom app for something specific, more people will do it. But they won't ask for an app, they'll just ask for a problem to be solved and the https://t.co/Av8xIVfNM9
  • 2026-10-15 — 2110797090763022346 — @threepointone the pipeline from agent doing a thing to deterministic software needs wheel greasing. This is what Kody is doing and codemode enables that. https://t.co/FcMNJHazpN https://t.co/zVBky3CLtZ
  • 2026-10-15 — 2110746987800994114 — @Paul_Kinlan @iabdul_wasey Typically merges don't happen this way. This is just a stress test for that. Normally the agent updates it's branch, runs the tests, and pushes
  • 2026-10-15 — 2110746661857468443 — @rtbenfield Always ahead of the game: - Testing Library - distributed data next to compute - MCP - Product Engineering - Invisible software
  • 2026-10-15 — 2110623631458062374 — @pid1_ I mean, the agent will get a regular git error when it tries to push a commit and it's not on the latest HEAD so it will rebase and then try pushing again. Doesn't "trigger everything again"
  • 2026-10-15 — 2110587245254365333 — @zeeg I actually have a Grok bot that I drop links that I think are AI-replies and it sends them a DM from @kodykoala telling them that I'm making a video about this and asking them why they're doing it and if it works. Most do not reply.
  • 2026-10-13 — 2110050828925259930 — @Elev_30 the agent never sees the token. It's proxied through Kody which replaces the token, but only if it's safe to do so: https://t.co/ZcR1CbHc2e
  • 2026-10-13 — 2110027468161921047 — @PAULIEWOOD AI reviewers have gotten really good! And if you have agents creating good PRs in the first place then the job is easier. And if you reduce the blast radius if something bad goes out then you feel more confident shipping.
  • 2026-10-13 — 2110027098152898665 — I found out today that the Kody codebase passed 1 million lines of code. That excludes self-hosted Kody and the Kody CLI. Probably a good time to send in the reaper agent 💀
  • 2026-10-13 — 2109886495725379694 — @poteto I don't like writing what I know other agents will read and I prefer talking out my thoughts and having the agent turn that into a coherent structure for me. But yeah, if humans will read it then I want to author it (though it's still voice to text lots of the time).
  • 2026-10-13 — 2109860735702425787 — Engineering away "slop" is one of the most fun things I do! I don't love the term "slop" because it frames it like this is a new thing and it's not. That said, preventing it, is a pretty simple 3 pronged process: 1. Ensure agents create high quality PRs in the first place 2. https://t.co/Q2Zs5Ew5IL

John Lindquist (@johnlindquist, id 15430177)

Adam Gospodarczyk (@_overment, id 1018768036373856256) — architect