Pramod Prasanth

Building production-grade AI for enterprise decision-making. Supply chain veteran who's done with vendor vaporware. Basel. Photography. Bikes. Emerging markets.

[Pramod Prasanth @PramodPrasanth]

@businessbarista @tenex_labs Hi Alex, if you ever look to expanding to Europe I'd be very keen to explore opportunities. Check https://t.co/h2k9Is6332

[Pramod Prasanth @PramodPrasanth]

The cockroach is a master of survival. But the one lineage that stopped surviving and started building became termites.

The movement has the name. The harder thing, the thing no boot can wait out, is to start building. (7/7)

https://t.co/umB1K8cOLA

[Pramod Prasanth @PramodPrasanth]

A judge on India's Supreme Court called unemployed young people cockroaches.

A generation took the slur and made it a flag. Twenty-two million followers later, the real question is what happens to that anger next.

I wrote them a guide. (1/7)

[Pramod Prasanth @PramodPrasanth]

The essay ships with a tool: a prompt that turns ChatGPT, Claude or Gemini into an interviewer, plus a printable question set.

You can run it at your kitchen table tonight. (6/7)

[Pramod Prasanth @PramodPrasanth]

@hthieblot Preflight

[Pramod Prasanth @PramodPrasanth]

A more capable model is a more willing one. It takes bigger steps on one instruction. Capability raises the size of the action, not the boundary around it. The harness is the cockpit. You check on the ground, where it is cheap, not in flight. 4/4

[Pramod Prasanth @PramodPrasanth]

Last week I asked an AI agent to change one function in one file. I came back from making coffee to find it had also edited three other files and a config at the project root. None of it wrong. All of it outside what I asked. There was no undo. 1/4

[Pramod Prasanth @PramodPrasanth]

In development this is cheap: version control, a red test, a reviewer reading the diff. In a regulated process the same slip is a deviation report, an audit finding, a paused line. Why the harness, not the model, is where enterprise AI breaks: https://t.co/2nAcDoLssc 3/4

[Pramod Prasanth @PramodPrasanth]

The honest question is never who has good values. It is what holds them when they do not. (4/4)

[Pramod Prasanth @PramodPrasanth]

Three AI labs. Three different responses to the Pentagon's autonomous weapons contract. Only one pattern explains all of them. (1/4)

[Pramod Prasanth @PramodPrasanth]

"Better me than the alternative" is the strongest argument in tech, and the one claim that can never be checked. You cannot verify what a person is really chasing. So stop trying. Watch the conduct. (3/4)

https://t.co/PyEiobQLi4

[Pramod Prasanth @PramodPrasanth]

I work in a narrower corner: regulated enterprise. There the human stays not because the model isn't good enough yet, but because the law requires a signature. That doesn't move when the next model ships.

A manager in Dan's frame. A signatory in mine. New essay:

[Pramod Prasanth @PramodPrasanth]

Dan Shipper on Lenny's Podcast this week: "automation is a lie. Every agent needs a human."

His company is as AI-forward as they come. It also doubled headcount in a year.

Most of what he predicts for the next year, I think he gets right.

[Pramod Prasanth @PramodPrasanth]

It's whether you can prove how a decision was made: on what data, with what authority, who could have intervened.

Most orgs built the model layer. Almost none built the layer beneath it.

I call it the proof gap. New essay:

[Pramod Prasanth @PramodPrasanth]

Three 2026 surveys. Three methodologies. One finding.

Grant Thornton: 78% can't pass an AI governance audit in 90 days.
Stanford: 88% adopted AI, under 10% scaled it.
Deloitte: 74% will deploy agents, 21% can govern them.

The bottleneck isn't the model anymore.