Pinned Tweet
I’ve officially submitted CivicFlow to the @OpenAIDevs WebMCP Challenge!
Having worked on a health care exchange portal for a US state, I know how much concentration these applications can demand. People may spend hours reading instructions, checking details, and carefully moving through every section.
That experience inspired CivicFlow.
CivicFlow is a synthetic public-benefits portal where a person and a browser agent work on the same visible application. Instead of making the agent guess what to click, WebMCP exposes typed, contextual actions for checking progress, navigating sections, and helping with household, income, coverage, documents, and review.
I also built an optional Gemini Live voice companion that uses the same WebMCP tools. Changes remain visible and reviewable, and the person stays in control, including the final human-only submission step.
This isn’t an eligibility engine or a real government service. It explores a bigger idea: could structured agent assistance make complex applications less mentally demanding and easier to complete correctly?
Would love for you to try it:
civicflow.codesm.chatgpt.sit…
Submission:
devpost.com/software/civicfl…
Devin was already impressive with SWE 2.
Now with Devin Bots coordinating teams of agents, things are getting even more interesting.
Really curious to see how well they handle task delegation, handoffs, and reviews in practice.
Lesgoo!
Meet Devin Bots.
From writing firmware to merging thousands of PRs a week, a Devin Bot leads a team of agents to get the job done.
1
17
I've been doing this for some time now. No need to get on the localhost anymore. And this is faster!
I haven't touched localhost in month.
I ask AI to make the feature only visible to me in production, so I can assess and iterate. If it's good, I release it to users, and ask the agent to set hourly/daily reminders to check logs and database to fix any issues. My AGENTS·md also has rules to force the agent to test everything.
Now AI is so good, everything goes straight from prompt to prod.
34
Google is surely going to sit on top with Anthropic in a few weeks.
And in my opinion, a lot of this got possible after the recent leadership change at Google DeepMind.
20
From Grok bot as Chief of Staff to a mobile as an AI Chief of Staff.
This is insane. Built on Google Pixel 10, it brings more trust!
Kudos to the team.
We switched 10 people from iPhone to Blue. Weeks later, none of them want their iPhone back.
Introducing Blue: a phone you just talk to. Tell it what you need, and it gets done.
It has all the context, can use every single app, and even triages your notifications.
We built it for people who always have a little too much to do. The ones buried in messages, notifications, and to-dos.
If that's you, get yours today.
Backed by Y Combinator + 70 other VCs & angels.
heyblue.com
28
Too lazy to apply for 20+ jobs a day or find time to automate that part?
Here's the best solution, in my opinion!
I love these types of connectors.
Introducing job search tools for ChatGPT & Claude.
→ Match your resume to 7M+ open roles
→ Find hiring managers to email from 250M+ people
→ See what they care about before you reach out
Run your job hunt from any harness with the Firecrawl plugin.
alexandria.firecrawl.dev/go/…
26
Theo's AGI test: GitHub stops putting the code above the README.
His ASI test: Google ships software that's pleasant to use.
Bro I'm crying 😂
44
My Devin Max seat ends tomorrow.
Got it free from @markfenner, and I used every bit of it.
SWE-2 ran my orchestrator agent and did almost all of my implementation work.
I built a few internal products on it that I'm finishing this month.
I tried the heavier options too. Fusion with Opus and Sonnet is great.
But SWE-2 at Max is still my pick for coding.
The workhorse that just gets it right.
And it's free in Devin till Nov 30, so I'm not slowing down.
Thanks Mark 🙏
2
2
93
SWE-2 has been my main builder for weeks.
A fast version is exactly what the small fix loops needed. Trying it this week.
Introducing SWE-2 Flash, our new fast & efficient coding model.
On leading evals, it outperforms all American open-weight models, even ones 3x larger.
To balance transparency and cyber risks, we are building a program to securely share the weights with partners & researchers.
24
My review agent kept flagging everything.
My coding agent kept trying to fix everything.
Most of it was stuff no user would ever hit.
So I added one rule.
Every bug gets two scores from 1 to 10:
how often it happens
how bad it is when it does
Multiply them. Under 5, it goes on the issue list and the code stays alone.
17
India doesn't speak demo English
Only 0.02% of Indians speak English as their first language.
About 10% speak it at all.
43% grew up speaking Hindi.
And most of us mix the two in one sentence.
Research shows speech AI makes 30 to 50% more mistakes on mixed speech like that.
That's why most voice AI demos fall apart on a real Indian phone call.
23
Nobody talks about this part of building things.
You spend 3 hours stuck on one problem.
The fix turns out to be one line.
To everyone else it looks like 5 minutes of work.
Most of the job is the part no one sees.
What's something you did that looked easy but wasn't?
1
22
Just finished Tibo’s game.
Hammer round. Absolute cinema. @poteto cooked.
Day 28/ tibo's game has landed
i spent $5.99 on this domain, but i had a blast making it with grok bot and cursor! i hope you enjoy it and find all the easter eggs 🤍
if i did it right you should be able to play it inside of X too
tibo.dog
41
Sometimes I type terrible instructions at my desk.
Half my best ones started as a 90-second ramble while walking.
Same agent. Same repo. Way clearer when I hit the mic first.
21
This is the playbook.
A few good bots beat a pile of vague ones. My CoS only talks to specialists that own one job.
here’s a short grok bot guide on how i create high quality bots and manage my team. ask @bot to read this post and help you improve your team!
• add dr eggbot to your team! he’s my bot designer that i use to create bots
• you don’t need a lot of bots. a few good bots can be more useful and easier to manage than a large team
• tell your bots to create skills for itself as you give it instructions on how to do tasks. this lets them do them more consistently even over long conversations. you can even ask it to create a private github repo for you to store all your skills if you want them preserved independently
• add connectors, use the vault, and give access to all the tools you use to do your work. you can also ask your bot to setup its own email and it can sign up for its own accounts for better separation
• thinking about all the work you do by hand, or steps in your workflows where you have to copy and paste information back to your bot. ask your bot how it can do that for you instead!
• imagine you hired an intern, how would you onboard them and help them do the work correctly and efficiently? introduce yourself to your bot and describe the work you do, then ask it how it can help you automate it
and my favorite tip: use voice mode or the transcription option! it’s a good way to just brain dump a lot of information. then, at the end, tell your bot to “restate this in your own words” so you can confirm it understood you
x.ai/bot/_jOdbfkB16zxu7MRcmR…
13
I've so much gratitude for Grok @bot and team.
It made my developer life easier and way more productive.
Most GOATED product I've used.
20
Google is shutting down Jules. Read-only Nov 11, gone Dec 31.
I used it early on. Weirdly fun, painfully slow, and a mix of solid code and slop.
My setup now: Devin for heavy lifts, other agents in isolated worktrees. I review and merge.
Tools disappear. My branches don’t.
It's been incredible watching you build, automate, and ship faster with @julesagent. Your experience has been instrumental in shaping developer products across Google.
The future of coding is just getting started 🚀🦑!
1
1
33
Cost per task + "Beyond the specs" is what I actually use when picking what runs overnight.
We've evaluated Gemini 4 Carbon and Gemini 4 Argon on FrontierCode 2.0, our coding benchmark that measure's a model's ability to go beyond the spec.
Gemini 4 Carbon scores near SOTA with a score of 40% and a cost of $4.04 per task, behind Claude Opus 5.5, GPT-6 Astra, and Claude Sonnet 5.5.
Gemini 4 Argon scores 26.3% with a cost of $6.41 per task, ahead of Muse Wonder 1.0, but behind Claude Haiku 5.5 and Grok 4.7.
See the leaderboard with full scores: cognition.com/frontiercode
25
Subhav retweeted
Replying to @SubhavMathur
terryhu.cn
minecraft-style personal site where you can actually walk through!
open source: github.com/terryHu7/homecraf…















