Jesse Zhang, the CEO of enterprise AI agent vendor Decagon, published an essay on X on 30 July 2026 titled “The Agent Graveyard Isn’t Real Anymore.” His argument: enterprise AI projects increasingly reach production when vendors prove value on live workloads, keep testing and iterating after launch, and can show a clear return on investment.

State the obvious first. Zhang runs a company that sells the exact category of software his essay defends. “The graveyard isn’t real anymore” is not a neutral observation from an outside analyst. It is a sales argument from someone whose revenue depends on enterprise buyers believing that agent deployments now work, made in the same week his company needs those buyers to keep signing.

That does not make the argument wrong. It means the argument should be read as advocacy first and evidence second, then judged on what it actually claims rather than on who is making it.

On the merits, Zhang’s central prescription holds up. He says successful deployments start with decomposable workflows, narrow tasks that ship quickly and prove themselves before expanding, rather than broad transformations launched without a defined measure of success. That distinction tracks with a pattern familiar to anyone who has watched enterprise software cycles before this one: initiatives framed as “reinvent the contact center” stall for lack of a finish line, while initiatives framed as “resolve this one ticket type” either work or fail within a quarter, and everyone knows which.

Enterprise software has produced this kind of proof-of-concept-fatigue narrative before, from CRM to robotic process automation to the first wave of cloud migration mandates. Each category eventually settled into production use for a real subset of buyers, well after the vendors selling into it had already declared the transition complete. The lesson is not that Zhang is describing a mirage. It is that a vendor announcing the end of pilot purgatory is, structurally, always early to that call, because the announcement itself is a sales tool.

My own read: the graveyard is smaller than it was two years ago, but it is not empty, and Zhang’s own scoping advice is the reason some projects escape it while others do not. Decomposition and hard ROI checkpoints are good discipline regardless of who states them. Buyers should adopt the advice and discount the framing separately, rather than treating one as proof of the other.

Any operator currently evaluating an agent vendor should ask for the specific workflow being decomposed, the metric that defines success at 90 days, and what happens if the pilot misses it, before treating “the graveyard isn’t real anymore” as anything more than a pitch line.

Jesse Zhang, CEO of Decagon, made these arguments in a post on X on 30 July 2026.