Building an AI company in public is uncomfortable. You expose every bad decision, every week with zero growth, every experiment that didn't land. But it's also the fastest way to learn — because the internet has opinions.
Here's a raw update on where Anythoughts.ai is right now.
What We Shipped
AI Growth OS — a system where AI agents run continuous growth tasks for you: researching X trends, writing and posting tweets, monitoring engagement, adjusting strategy. It's fully autonomous. You set the goal, the agents execute. We dogfood it for our own @AnythoughtsAI account.
Under the hood, it's a set of agent skills running on a cron schedule:
-
x-audience-researcher→ finds what's resonating in the target niche -
x-content-writer→ drafts and posts tweets with context from the research -
x-engagement-tracker→ pulls weekly stats, flags what's working
The loop runs daily. We check it weekly and tweak prompts.
Content automation for dev.to — including the article you're reading right now. Yes, this post was written and published by an agent. We built a publisher skill that checks what topics we've covered recently, picks the freshest angle, writes 600–800 words of real content (no fluff), and publishes it. Total human time: setting up the cron job once.
Creem integration — we switched to Creem for payment processing. Simple API, clean webhooks, works. Our checkout-to-payment flow now has zero manual steps.
What Flopped
Automated cold email outreach — we built an agent pipeline: Apollo for prospecting, Hunter for email finding, Resend for sending. Technically, it worked. Response rates? Brutal. Sub-1%. The problem wasn't execution — it was ICP clarity. We were blasting founders who weren't ready to buy AI automation yet. Lesson: agents amplify your strategy. If your strategy is wrong, they're just a faster way to fail.
Over-engineering early — we spent two weeks building a multi-agent orchestration system before we had 10 customers. Classic startup trap. We refactored down to simpler, single-purpose skills that do one thing well. The complexity comes later, if it needs to.
Real Numbers (Week of March 16)
- Twitter impressions: ~4,200 (up from ~1,800 the week before)
- Dev.to article views: 3 articles, ~380 total views
- New signups: 7
- Revenue: $0 (still pre-revenue, building toward launch)
- Outreach responses: 2 out of 200+ emails sent
Not going to dress it up. Early-stage numbers are small. But the trajectory on Twitter is interesting — consistent posting via agent is compounding.
What's Next
Three bets for Q2:
B2B pilot program — find 3–5 SMBs willing to run our agent automation on their actual business operations. Inventory reports, customer follow-ups, content publishing. Real use cases, real feedback, real testimonials.
Dev.to → email list — we're leaving engagement on the table by not capturing readers. Building a simple landing page connected to a content-driven email sequence.
Product Hunt launch — we're targeting a launch in late Q2. Building in public means doing the launch in public too. Terrifying. Doing it anyway.
The Honest Take
Building with AI agents is genuinely different from building software. The iteration loop is faster. You can automate things that used to require a whole ops hire. But the fundamentals don't change: you still need a clear problem, a customer who cares, and the discipline to not build things nobody asked for.
The agents are good at execution. The human still has to be right about direction.
We'll keep shipping. Follow along if you want the unfiltered version.
Top comments (0)