The biggest disparity I’m seeing right now with AI is the gap that exists between enterprise adoption and the rest of the broader market narrative. There is absolutely no doubt that engineers at very large enterprises have been exposed to coding agents or experimented with them in their personal lives. But the gap that exists between these larger enterprises truly adopting agents in their SDLC and also seeing real productivity gains from those investments is way larger than most people realize. This morning I had a conversation with a CIO of a large public enterprise with 10k+ engineers where they’ve only just now kicked off a POC with a few different coding agents. Another one of our larger customers actually rolled out Claude Code to their engineers and pulled it back after seeing a huge spike in incidents. The reality is that larger enterprises have enormous foundational requirements that they have to first get a handle on before exposing more of their SDLC to AI. Guardrails have become a must have and not an afterthought in a Confluence doc or spreadsheet. And coding agents are still part 1 of a longer transformation in building the software factory that I suspect will take a few years to truly materialize. This transformation that is happening in the enterprise is not unlike the on-prem to cloud migration that has taken better part of a decade and still happening across the industry!
Boom 💥!! Love your POV Anish Dhar AI adoption & Agentic AI for large orgs with complex tech stack spanning over decades is complicated. The playbook is still in the making. The Claude Code pullback says it all, Anish. And that incident spike was AI exposing guardrail gaps at agent speed. The orgs treating quality as shift-left and observable absorb this. The ones treating guardrails as an afterthought get the incident graph. Gate in the pipeline and governance are key for success.
Manish Yadav
4d
The distinction between AI experimentation and AI-enabled productivity is critical....Giving engineers access to coding agents is the easy part; building the guardrails, workflows, security, and engineering foundations to use them safely at scale is where the real transformation happens.
The gap between exposure and adoption is easy to miss. An engineer using Claude Code on their own laptop is very different from a 10,000-person engineering organisation giving an agent access to production workflows. The second one changes the risk, governance, and infrastructure requirements completely. Maybe that's why the enterprise transition is going to look much slower from the outside than the capability curve suggests.