Rohit

AI agents that ship production software | Co-founder @ Yanflow | Built AI products for 500M+ users | IIT Kanpur · ISB

[Rohit @imrohitkgupta]

The biggest unsolved problem in AI isn't intelligence.
It's trust at machine speed.
Agents can now operate 100x faster than any human team. But the moment you bring humans in to supervise — you've throttled the only advantage that made agents worth building. Take humans out — one bad run breaks a customer, a compliance boundary, a business.
Nobody has cracked this handshake. Everyone is guessing.

But here's what most people miss — we're solving this with the wrong mental model. We keep asking how to fit agents into organizations that were designed for human speed. That's the wrong question.
The organizations that define the next 20 years won't look anything like today. Fewer people. Faster cycles. Agent networks carrying institutional memory and shipping — with humans setting direction, not managing execution.

That future is closer than anyone is ready for.
The infrastructure to run those organizations doesn't exist yet.
That's what we're building at Yanflow.
Not for the org of today. For the one that's coming.

[Rohit @imrohitkgupta]

@vahidf24 @Evalyze_ai Hi, hope this message lands

[Rohit @imrohitkgupta]

Outcome-based pricing sounds simple: charge for the result. Then you have to prove, in real time, the result actually happened.
Counting tokens is straight forward. Counting outcomes means building an attribution system before you're allowed to charge for one.
No wonder the big labs are charging for tokens, not results.

[Rohit @imrohitkgupta]

Buying software used to mean feature comparisons. Now it means asking an AI what it recommends.

I made two purchase decisions this month without really making them.

For email, I asked Claude how to move faster. It skipped Mailchimp (assistant mode) for Resend (it can actually run the workflow via MCP).

For ads, Google Ads has no native Claude access. Claude pointed me to Adspirer instead — full read/write, straight through Claude.

Both times I asked “how do I save time.” The AI made the call.

The moat used to be the tool. Now it’s how little of your attention it needs — and whether the AI puts you on the list at all.

[Rohit @imrohitkgupta]

This hurts. PM burnout: 44.7% → 55.7% in a year.

AI tooling didn't remove PM toil, it moved the bottleneck from writing slowly to reviewing fast, and reviewing under time pressure is its own kind of tired.

[Rohit @imrohitkgupta]

Intercom's Fin: $1M → $100M+ ARR at $0.99 per RESOLVED conversation, with a $1M miss guarantee.

Outcome pricing is a reliability disclosure. You can only charge for outcomes you control.

Most AI products can't afford to be priced honestly. Yet.

[Rohit @imrohitkgupta]

Copilot's pitch: lean on agentic workflows. Copilot's new billing: a single agentic session costs 3–4x your old monthly plan.

Pricing that punishes the behavior the product encourages isn't a pricing bug.

It's a broken value theory - charging for what you spend, not what users gain.

[Rohit @imrohitkgupta]

@vaibhavbetter Absolutely, we’re seeing the same in our customer conversations. Preference to make people faster using AI not transfer the disproportionate risk to executives.

[Rohit @imrohitkgupta]

Gartner: 40% of agentic AI projects will be cancelled by 2027.

The failure modes: 52% poor data quality, 21% governance maturity, 41% ROI within 12 months.

It's not a capability problem. It's a governance problem.

Quality built into generation. Not reviewed after the fact.

[Rohit @imrohitkgupta]

OpenAI: $13B revenue, 300M users. Delaying its IPO because $1T is too rich for public markets.

If the most iconic AI company can't go public on its own terms, every AI founder's next round will ask: when does revenue catch the narrative?

The "figure out monetization later" playbook seems to be dying in public.

[Rohit @imrohitkgupta]

Anthropic is having to introduce mandatory team lunches and hackathons because going all-in on AI agents made their engineers' work feel lonely.
If the most AI-forward team in the world needed organizational interventions, most companies have zero plan for the collaboration layer AI execution is quietly dissolving.

[Rohit @imrohitkgupta]

@kushalbhagia Wow, to make it even more interesting, one can opt for annualised agent runs per dollar. that would surely make their CFO happier.

[Rohit @imrohitkgupta]

@ashwingop Sadly, not many understand this trap until it’s too late!

[Rohit @imrohitkgupta]

SpaceX earns $2.32B/month from AI compute and Anthropic, Google, Reflection AI all paying rent.

A rocket company is becoming the AWS of frontier AI infrastructure.

The model race is real. But actual game is in owning the data center.

[Ksenia Moskalenko @kseniam0s]

FOUNDERS: You've been paying to build on Claude. @AnthropicAI launched a program to change that.

@Claudeai for Startups - free API credits and priority rate limits for early-stage VC-backed founders:

- Free Claude API credits
- Highest rate limits, no throttling in production
- Hackathons, Founder Days, and meetups
- Early access to new model releases

Build with the full Claude stack:
Claude API, Claude Code, Claude Managed Agents, and Claude Cowork.

To qualify: your startup must be early-stage and backed by one of Anthropic's partner VCs. Ask your investors for a unique application link.

Apply → https://t.co/HckO4O93Ho

P.S. Founders using Claude to build - when you're ready to raise, @ThePageform is where your data room lives → https://t.co/RgOL0J0kt5

[Rohit @imrohitkgupta]

@RamaswmySridhar @Snowflake Thanks for this. Makes a good case for comparison per jobs.

[Rohit @imrohitkgupta]

Google paid $2.7B to retain one researcher. He left in 18 months.

7 of 8 Transformer authors have now left Google.

Money isn't the retention mechanism for the people who matter most. Problem quality is. When the hardest work moves somewhere else, so do they.

[Rohit @imrohitkgupta]

It baffles me that the company best positioned to bring AI natively into everyday workflows is struggling to make its frontier models indispensable.

I’ve tried Gemini repeatedly, but it still hasn’t become part of my daily workflow. The reach is unmatched. The utility isn’t, at least not yet.

Not sure what finally changes that.

[Rohit @imrohitkgupta]

@dessaigne Agreed in principle. But savings are a guaranteed outcome; productivity is a claimed outcome. Buyers know exactly what they save, but not necessarily what they gain from shipping 3x more.
Until output is tied to revenue, growth, or margin, cost remains the easier story to buy.

[Rohit @imrohitkgupta]

Windsurf rebranded to Devin Desktop on June 2.

The shift: "code-first IDE that can call an agent" → "agent management hub that contains a full IDE."

They're betting orchestration and governance are the durable value. Not autocomplete.

The IDE war and the agent war just merged.