I remember when building a software product meant hiring a team. Backend engineers, frontend developers, a designer, a project manager. Even as a lean indie founder, you needed at least 2-3 people to ship something real.
Today I'm running 14 products with an agent that works while I sleep.
Last week, JPMorgan Chase, American Express, and Red Hat joined the Agentic AI Foundation (AAIF) — a Linux Foundation initiative anchored by Anthropic's Model Context Protocol (MCP). When institutions like these formalize their involvement in the infrastructure layer of AI agents, something fundamental has shifted. This isn't experimental technology anymore.
But while the enterprise world catches up, indie founders have already figured this out. Here's what's actually working.
MCP: The Protocol That Made Agents Real
If you've been building with AI and haven't internalized MCP yet, stop and read this carefully.
MCP hit 97 million installs on March 25th — the fastest-adopted AI infrastructure standard in history. OpenAI, Google, Microsoft, Amazon — everyone adopted it. But more importantly for us indie builders: it's the protocol that lets AI agents actually do things.
Before MCP, AI was a conversation partner. You'd ask it something, it would answer, and you'd go do the thing yourself. With MCP, the agent connects to your tools, reads your data, takes actions. The gap between "AI advises you" and "AI does it" collapsed.
At Inithouse, our whole operation runs on this. Claude Code with a set of custom Skills handles our backlog, monitors SEO, writes and publishes content, checks metrics. It works on scheduled tasks — Anthropic calls it /loop — running in the background like a junior employee who never needs sleep.
The Loop That Changed Everything
In March 2026, Claude Code shipped the /loop command — scheduled background tasks. Combine that with Computer Use (Claude can now click through interfaces, just like a human) and what you have is genuinely autonomous work.
For context: one of our products is Watching Agents — a platform tracking AI agents across categories. The irony isn't lost on us that we're using agents to monitor agents. But it's real: agents doing research, agents updating databases, agents writing first drafts.
The critical insight: you're not just automating tasks, you're multiplying your surface area. One person can now maintain presence across 14 products simultaneously because agents fill the gaps between human decisions.
This article you're reading? It was researched, drafted, and published by an agent running a scheduled task. I reviewed and approved the direction. The agent did the legwork.
Lovable Changed the Build Cost Equation
A year ago I would have told you the bottleneck is development time. Then Lovable hit $400M ARR — up from $200M at the end of 2025, achieved with just 146 employees. The product hit that growth by doing one thing right: it made shipping a real app something any founder can do in a day.
Every product in our portfolio — from Be Recommended (AI visibility tracking for brands) to Vibecoderi.cz (the Czech vibecoding community hub) — was built in Lovable. The stack is React + Supabase, published in minutes.
One thing I'll flag: Lovable-generated apps recently had a security audit that found RLS (row-level security) misconfigurations in ~10% of surveyed apps. This is the cost of moving fast. Be explicit about security in your prompts, and audit Supabase policies manually before launch. Fast building isn't an excuse to skip basics.
The Real Bottleneck Now Is Distribution
Here's the uncomfortable truth I keep coming back to: the tech barrier is essentially gone.
If MCP + Claude Code + Lovable can build and maintain 14 products with one founder, then thousands of other founders are doing the same thing with the same tools. The products themselves aren't the moat anymore.
The data confirms this: AI-powered indie founders reach PMF 2.5x faster, but in a market where micro-SaaS is growing at 30% annually and heading toward $59B by 2030, distribution is the only sustainable advantage.
This is where building in public matters — not as a buzzword, but as a genuine acquisition strategy. When you share what you're learning — the actual mistakes, the agent prompts that failed before they worked, the metrics you measure — you build something competitors can't copy: accumulated attention from people who trust you.
What I'd Tell Myself 18 Months Ago
Invest in your agent stack early. The time you spend learning MCP, structuring tasks for agents, and building reliable scheduled workflows compounds. A month of learning buys you years of leverage.
Build on platforms on an upward trajectory. Lovable going from $200M to $400M ARR in 6 months tells you something about where the ecosystem is heading. The tooling will only get better. Bet on momentum.
Distribution channels are the product. I spent months perfecting features that nobody saw. Now I publish what I'm learning — in posts like this — and build audience alongside product.
The agents will do more than you expect. When we started delegating work to automated agents, I expected 60% quality. I got 90% with iteration. The ceiling keeps moving.
What's Next
The AAIF MCP Dev Summit runs April 2–3 in NYC. The fact that a dedicated conference for an AI protocol exists — with 146 corporate members behind it — tells you this isn't a trend, it's infrastructure.
For indie founders, the question isn't whether to adopt agentic tools. It's how fast you can build the operational systems that use them effectively before the next wave of competition catches up.
We're still early. But early is ending.
I write about building AI-native products as a solo founder at inithouse.com. If you're experimenting with agent stacks, share what's working — I read every comment.
Top comments (0)