Jonathan Corbin

On stage at @RaiseSummit in Paris this week. The real conversation in enterprise AI has moved on from models. Companies are now asking: “How do we actually get agents into production - reliably, at scale?”That’s the problem we’re solving at @MavenAGI .

Exactly what we’re seeing in the enterprise. The emergence of on prem as a measure of protection against unscrupulous vendors who are using your data and user interactions to train their models.

Some airlines are great at making you feel like a welcome guest. Others are exceptional at expressing their indifference.

👀🍿🍿🍿

We’ve come a long ways since every application layer ai product was just a “wrapper on a llm”

Quoted in Forbes from RAISE Summit this week: enterprises are done handing proprietary data to frontier models without control.

On stage at @RaiseSummit this week talking pilot to production. The shift is real: regulatory clarity turned enterprise AI hesitation into procurement.

Gartner: $234B of SaaS spend exposed to agentic arbitrage. The interface era is ending.

K1x's first bot resolved 15% of tickets. On Maven, 80%. Same model. The ceiling was never the model, it was the operating layer around it.

At Maven AGI the gap from 40% to 90% autonomous was never the model. It was the operator.

RAG tells an agent where to look. A Graph of Record tells it what's true.

A $3.6B acquisition makes the buyer bigger. It does not make the agent reach production.

A smarter model doesn't fix a context problem. It just guesses faster.

Everyone can demo an AI agent. Almost nobody can fix one in production. That's the whole game.

You can't coach what you can't see. Observability is what keeps an AI agent alive in production. Not a smarter model.

Clio tested 32 AI vendors. The winner didn't answer best. It resolved end to end.

Clio tested 32 AI vendors. The number that decided it: resolution under ambiguity, not the demo.

Microsoft says the seat is dying. Consumption is the half-step. Outcomes are the destination.

The average enterprise runs 12 AI agents. Half can't talk to each other. That's the whole problem.

🎯