I coded myself out of coding: and built a plain-English agent builder

Most of us jumped on “AI productivity” thinking it would reduce the grunt work.

Instead, we now juggle a maze of tools over hundreds of tabs.

AI was supposed to free us from busywork, and clear out our to-do lists, but faster outputs meant more time reviewing. more fragments to stitch. more apis and mcps to wire, zaps, more context switches.

I hit that wall while running growth ops. I’ve spent the last decade wiring marketing stacks and automations. LLMs gave me instant drafts and it felt amazing, but the “glue” work, prompt chaining, zap setups, troubleshooting, only got heavier. The answer to all this inefficiency was to better and smartly leverage flows, and agents, so I could offload the intermediary steps.

But that meant coding from scratch, langchain tutorials, or lean on pre-LLM "no-code" platforms, but that felt like the retrofits, and both paths actually added more friction and cumbersome to deal with. So I built Metaflow AI. My attempt to lighten the load.

Hey, I’m Narayan 👋 solo founder of Metaflow AI.

I’m not a VC-backed rocket ship—just a solo founder shipping, learning, and rewriting the spec each week based on real usage. If the idea of talking to your agents and workflows in natural language resonates, follow me to stay tuned on Metaflow AI's official launch.

What it does in one sentence

- simply type instructions for your agents. click to connect any app. or simply '/' to add an action

- skip nodes, skip workflow designs, skip elaborate plumbing, just type short steps

- want to expand the context and capabilities for your agents, you still can wire visual workflows for reliable outputs that you agents can call without depleting its token limits on the main convo

Curious: what’s the most annoying multi-step task you’d hand off to an agent if English were the only interface?