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…

I gave an AI one week to handle my founder chores. It ended up running my coding agents. Here’s what I got @hark_labs to do: → Ran Codex CLI across 2 accounts, with live quota tracking on my home screen. → Launched Devin Fusion jobs in the cloud. → Worked with T3 Code on my Mac to coordinate my coding agents. → Repeated build → independent review until APPROVE, capped at 3 rounds. → Created branches, merged PRs, and pushed changes only with my approval. → Flagged failing GitHub Actions and upcoming hackathon deadlines. And somehow, in between all this, it checked me into a hackathon, saved my AWS credits, and cleared 43,000+ promotional emails. The best part? I can just text it from my phone while the coding agents keep working. I still approve the important decisions, but I no longer have to constantly jump between tools to coordinate everything. And the UI genuinely makes the whole experience enjoyable. Personal AI agents are getting seriously interesting. I’m going to keep pushing Hark to see how much more it can handle.

Update: Day 1 with Devin Max, and fusion mode surprised me. - Opus 5.5 (high) as lead - SWE 2.0 (medium) for execution Hours of quality work, only 15% of my weekly quota used. @cognition, let Cloud Desktop pick models per fusion setup and this is 100%. What combo are you running?

This is something I wouldn't want to miss. I'm speechless at the moment, tbh.

Just won a free month of Devin Max! 🚀 Huge thanks to @markfenner and @cognition for making this happen! Planning to put it through its paces on real projects, compare it with my current coding agents, and regularly share what works, what fails, and what surprises me. Excited to try it out!

A random thought at 3 AM. There are some incredible people out there doing things I genuinely admire. And I know I’ll work alongside them someday. Not just one, but many. I don’t have those connections today. I haven’t earned my place in those rooms yet. But I will. I’m going to get hungrier. Build more, learn more, put myself out there, and earn my way in. See you on the other side.

This will be life-changing. Giving personal assistant bots automatic access to better models based on the task is a step towards AGI. But I wonder what the cost impact will be on usage.

Got off the waitlist for @OpenAI DevDay Bengaluru 🎉 The bad news: I got in so late that flights are now ridiculously expensive. Talk about good news with terrible timing :(

Time to update my specialist bots. My favourite is my fleet interviewer, which builds a team for any product I'm creating. What's yours?

This is a true personal AI. Nothing leaves your storage. All data is private. So damn impressive!

The $20 plan has no business feeling this powerful. GPT-6.1 Sol High for implementation, Max for review, and Ultra for planning has been an insanely good combo for me. And if you pair it with something like Cursor or Antigravity, the whole setup starts feeling almost unlimited. For the price, the capability + reasoning + code quality is kind of ridiculous. Goated setup. 🔥

Time to use my stale 3 Pro accounts. 😙

I realised I was wasting Grok Bot quota by making it do too much. So I split the workflow properly. Grok Bot handles the management side: Chief of Staff → Engineering Manager. Once a task is scoped, it gets handed to a Cursor Cloud Project, which orchestrates the actual engineering work: Codex plans/reviews. Cursor Agent implements. My Mac provides the execution environment. Then the result goes back to the Engineering Manager. Much cleaner boundaries, much less Grok Bot usage, and I can trigger the whole workflow from anywhere.

If Gemini 4 ends up anywhere near GPT-6.1 Sol pricing while staying in Opus/Fable-class territory, multi-model architectures get a lot more interesting. Model councils. Swarms. Independent critics. Parallel agents. Suddenly, running 5 frontier models on one problem doesn’t sound ridiculous. Intelligence is getting cheaper while staying incredibly capable. Awesome time to build. ⏫️

Google AI is finally making moves. Benchmark-wise, they're crushing them. Waiting for their general-purpose model release near Fable and Astra level. I have 3 Gemini Pro accounts. Finally seeing some hope in using them again. 🥹

Oh I was so wrong. Nothing for plus people. Hope Sol 6.1 can compensate.

As far as I can tell. Plus users will benefit the most? In terms of expectations ofc.

Cleared my schedule for @OpenAI DevDay. Cute ASMR animation playing while I wonder how many SaaS startups are about to get butchered today 💀

This is world-changing. It gave me goosebumps yet I felt terrified. Huge opportunities lie ahead. Fucking love voice agents!

If this rolls out, I’m not sure I need it. I already have Cursor Pro+ and X Premium, and I’m pretty happy with the pricing. A unified plan sounds convenient. I just hope it doesn’t become an expensive bundle for people who only need part of it.