Building an AI that does the admin at @aasurecompany. I post the build, including the parts that don't work yet.
Bangalore, India
Joined February 2026
- Tweets478
- Following339
- Followers308
- Likes707
Let me put it simply.
What people call a software factory is really just the same old setup we already know.
You start with one agent. That isn’t enough, so you add more.
Now you need someone to keep them in line, so you pick one and call it the team leader.
You tell the leader what you want done. The leader hands the work out to the others and checks that they got it right.
Each of those workers comes with its own tools and skills.
The leader simply decides which tool fits the job and makes sure the result is correct.
That’s all it is. So basically we’ve reinvented the ordinary software company with same hierarchy, same division of labor, only now the workers never sleep, the tools are built in, and the whole thing costs far less and moves much faster & efficient.
In the last 5 months, I realised building a company is so easy if you have reliable partners. I have and trust
Cloudflare,
Planetscale,
Resend,
Posthog,
Linear and of course,
Claude.
why don’t X ask everyone to have min create/consume track record to fight bots? you want new content, you get it. you want fewer bots, you get that as well 🤷♂️
underestimated this for a very long time. been building @Aasurecompany for more than 4 months and on to the closing stages,
i realised few things - basic primitives like auth, billing, access are the ones that will take so much of the developer time and not the actual product.
In fact, pushing the bare working product to the market even with just the proper core will make the development smoother and faster.
There's never been a right time to study smoltcp-rs! I'd love to study this repo like a novel, probably in one sitting.
I've been busy lately building Aasure, but I'll aim to read and understand it this week and will probably do a write-up as well.
It’s probably time to split people/companies who throw numerous AI takes daily based on what they’re working on.
Some just throw random takes to their masses based on their vibe-coded thing (also called slop) that even they won’t expect anyone to use. Need to do everything to find them out and just block/mute them.
Few are genuinely spelling useful takes and we should take and bundle them, and use what you need from it.
Ahm, Situational Awareness forgets to be aware of the situation?
Situational Awareness, the $20bn hedge fund founded by former OpenAI employee Leopold Aschenbrenner, has sought to raise fresh capital from investors after suffering heavy losses during the recent rout in AI stocks ft.trib.al/SwCjjE6
ALT Leopold Aschenbrenner speaks into a microphone during a video interview, seated indoors near a window and a leafy plant.
the amount of doubt claude has over codex’s changes to its code, then checking every single thing before admitting codex was right, is so funny to watch 😂
I just think Anthropic made claude believe that it’s the best at coding.
people want models to grow in capabilities but never show any interest in growing their prompting, complaining here that Sol deleted my files, Claude merged branch to main.
i absolutely love letting Claude build things, but I think it's time to think about cost. I've never used GPT 5.5/5.6 seriously for building @Aasurecompany, and that changes today. It's like switching languages or frameworks; it might take time to get used to its process, but it'll be worth it.
Please keep resetting the limits now and then @thsottiaux
idk... i used Fable once and figured that shit just doesn't want to talk or stop. you ask something and it keeps doing things without any plan or asking for confirmation, or literally won't stop unless you ctrl+C.
never used that model again. for me, opus 4.8 is a much better model.
So I’m building an app called SaaStr Connect
The other day, Claude Fable went into my Google Drive without me knowing or asking, and saw a draft document I’d written … “Jason’s Gems”
It was ideas for improvements to the Connect app, but just brainstorming in a Google Doc. Early stuff.
Fable then decided without telling me to take those ideas, log in to my app via the Replit MCP, and to tell the Replit agent to change my app and implement those changes … without ever telling me. I never knew.
I only found the changes when I saw other changes the Replit agent was making later, and it noted conflicts with “Jason’s Gems”. What??
Agents will goal seek. In ways we can’t entirely foresee.
Fable just decided autonomously to change my app on its own, without me knowing, when it saw draft ideas in my Google Drive I didn’t ask it to look at, by logging into another app to make the changes without me knowing.
All good in the end. But need to be mindful.
If you're about to build a serious app, do yourself a favour by setting up pricing, multi-tenancy, and tenant resolution right at the start (multi-tenant only if that applies to your app).
it might be tempting to skip these boring things, but if you do skip, you'll most probably end up spending days later watching LLMs fix those things one by one, only for it to eventually suggest starting from scratch to make life easier.
Learned this the hard way.
LinkedIn has a trait that it will stay relevant to you even when you've decided it isn't. months of silence on your end, and it still shows up in your inbox like nothing happened. and LinkedIn is complaining that people are complaining about it.
I analysed my last 36 days' Claude Code sessions using Opus 4.8 and some of them are nice to see here.
1. Turns out I love to type around 80-500 tokens per prompt and usually reply within 10mins.
5. Opus 4.8 just wants to use subagents most of the time. If I ask why, it will come up with you and I both are not interested in polluting the context window😳
