Most enterprises do not have an AI strategy problem. They have a builder problem. Leadership has already decided that agents should touch support, finance, engineering, and the messy middle of operations. They have model access. They have a slide that says agentic. What they do not have is a team that can turn that ambition into software that runs on Tuesday without a founder in the room.
That gap is easy to miss from inside a coastal lab, where the default assumption is that every company has a staff engineer who will wire the tools over a weekend. Most companies do not. Their best engineers are already committed to the product that pays the bills. Their operators know the workflow cold and should not have to become a second engineering team to get an agent into production. The result is a pile of pilots that never become a system. Call it the operationalization gap. It is not a model problem. It is a staffing and software problem wearing a model costume.
The useful reframe is blunt. All new software is agentic. Helping a company build agents is not a side practice next to software. It is software. The primitives are the same: a defined job, context, a loop, permissions, a review surface, and an owner. If you cannot name those, you do not have an agent. You have a chat window with a budget. Anthropic describes agents as models using tools in an environment-feedback loop, with checkpoints for human feedback.[1] Treat agents as software rather than prompts, and the work still has to be specified, tested, and kept alive after the kickoff call.
This is why the winning motion for the next few years is not another horizontal copilot pitched at people who already know how to build. It is showing up where the ambition is real and the bench is thin, and building the first agents as production software. Context in. Background execution. Reviewable output. A human who approves. Repeat until the internal team can own it.
If your plan assumes every buyer already has a team that can assemble this, you are planning for a market that is smaller than the one with the budget. Go where the ambition is funded and the building capacity is not. That is the work.
Key takeaways
- Most enterprises already have AI ambition and model access, and still cannot turn that into software that runs without a founder.
- All new software is agentic, so building agents is software engineering, not a side practice next to it.
- The near-term opportunity is where ambition is funded and internal building capacity is thin.
FAQ
Is the enterprise agent problem a model quality problem?
No. Model access is already common. The missing piece is a team that can specify a job, wire context and permissions, and keep a reviewable agent alive after the pilot.
What does it mean that all new software is agentic?
New systems increasingly do work through agent loops rather than fixed screens and scripts. Building those systems still requires specification, testing, ownership, and review. The form changed. The discipline did not.
Sources
Related Essays
The Operationalization Gap: Where AI Demos Go to Die
The gap between an AI demo and an AI deployment is called software engineering. Most organizations are not equipped to close it, and that is where all the value lives.
Run the Chief of Staff Agent on Yourself First
Executives who have not run a chief of staff agent on their own work are not ready to specify one for the enterprise.
The Budget Meeting Arrives Before the Operating Model
Seat counts and chat turns will not survive a budget review. Merged, reviewed work is the only AI coding metric finance can fund.