Product Manager by profession, building at the intersection of AI, product, and growth.
Writing about agents, product strategy, and what actually ships.
Most teams still deploy a staging build just to ask, “What do you think?”
Shiproom lets anyone review your Localhost, annotate the UI, and turn feedback into agent-ready tasks.
One command. No deploy. Closed alpha is live.
https://t.co/3UpM7A6Oci
I'm sorry, Jev can WHAT???
Image classification. Object detection. Bounding Boxes
No manual annotation. No custom training. No fine-tuning.
Potentially under 1 paisa per image.
Does this kill deep learning? No.
But it might kill the need to train custom models.
I waited 2 years for this day.
1 year of research.
4 months of making my own algorithm.
8 months of creating the product.
and now, 1 month of experiment to validate my hypothesis through https://t.co/HI3EG0YB6f
Zugzwang is a prediction market that runs like a social network for rational debate.
@zugzwangworld@boardyai , what do you think about this. Isn't this crazy. I think some chess grandmaster or someone at the same level should see this. Don't you?
Hello everyone,
For two years, I have been writing a research paper called Zugzwang.
It proposes a new socio-economic world order for how we interact with knowledge and money.
The research takes the form of a tech product built to test my hypothesis.
It begins with an experiment I will play with the world for one month.
I am launching the product tomorrow - 5th October 2026.
— Hrishikesh
PS : I made this film as an invitation to the world to join the experiment.
@MonodropApp Appreciate this, Monodrop. Shiproom is still early, but the thesis is simple: feedback shouldn’t die inside screenshots and chat threads.
Glad to see it on the shelf. Now back to shipping.
Exactly. The screenshot is evidence. The context packet is the handoff.
The goal is that the next developer or coding agent shouldn’t have to reconstruct what happened. It should be able to pick up the issue with the route, element, viewport, browser state, failed requests and reviewer intent already attached.
That’s the bet behind Shiproom.
AI can write the code.
The handoff back to AI is still a screenshot.
That gap bothered me enough to build a product around it.
Coding agents can generate and modify software at extraordinary speed. But the moment another person reviews the result, the workflow collapses into a Slack message, a Loom timestamp, a ticket, or “the button on settings is broken.”
Then the developer has to reconstruct the route, the exact element, browser state, viewport, failed request, and what the reviewer actually saw. After that, they translate the whole thing into another prompt.
That felt structurally wrong.
So I built Shiproom.
npx shiproom live
It opens a private review URL for your locally running app. A reviewer uses the real interface, clicks or annotates the problem, and leaves feedback. Shiproom captures the context around that moment and turns it into structured Markdown or JSON. Through MCP, a coding agent can read the feedback directly and update its status as the fix moves.
The tunnel is only transport. The product is the context packet connecting reviewer, interface, developer, and agent.
I've put more thought, engineering, and iteration into Shiproom than anything I've built independently before. It spans CLI tooling, real-time gateway infrastructure, deterministic capture, private review flows, exports, and an agent-facing MCP layer. Most of that complexity is hidden behind one command. That's the point.
AI made producing code cheaper. I think precise feedback is becoming more valuable.
Shiproom is my bet on that shift.
If you build with Cursor, Claude Code, Codex, or another coding agent, try it on something real.
npx shiproom live
https://t.co/2t30j1IZDD
Tell me where the workflow breaks. That's the feedback I care about.
Contrarian take: the future doesn’t belong to AI agencies building chatbots, voicebots, or copilots.
Those are features.
The real opportunity is AI-native transformation.
Take a company. Tear down how work moves through it. Rebuild operations, decision-making, customer experience, and internal workflows around AI from the ground up.
Not “Where can we add AI?”
But:
“If we started this company today, with AI available from day one, how would we build it?”
The agencies that can answer that question will define the next decade.
I’m an AI Product Manager.
By day, I take legacy systems, messy operations, manual workflows and turn them into AI-native products and systems that actually work.
I’m obsessed with the gap between what AI can do and what people actually adopt.
By night, I build.
Products. Experiments. Weird ideas. Things that test how people behave when you put something new in front of them.
Some will work. Some will fail spectacularly. I’ll probably talk about both here.
I’m on X to meet builders, AI people, product thinkers, founders and anyone equally obsessed with what happens next.
If that sounds like you, say hi.
I’m just getting started.