Agent Harnesses Are Dead. Long Live Agent Harnesses. João (Joe) Moura Apr 14, 2026

Agent Harnesses Are Dead. Long Live Agent Harnesses. I said it at a Dev Day 2025 (DeepLearning.AI conference) last year: Frameworks are cheap. A few people in the audience looked uncomfortable. But I think the statement aged well. Now it's not just frameworks. Harnesses are getting the same treatment, and the cycle is only getting faster. Building anything is getting cheaper by the month. Vibe-code an app over the weekend. Spin up an agent with a few API calls. The distance between idea and working prototype has collapsed to almost nothing, and will collapse to nothing faster than most expect. And yet the industry is spending most of its energy debating what to call the building tools. We built our harness back in 2023 when we launched, what people now all agree to be the right abstraction, of multi agentic opinionated agent abstractions and eventually exposed its internals (the framework, a.k.a. CrewAI Flows). The combination of Agents, Crews and Flows, allowed for a massive array of use cases and adoption across major enterprises. We have seen the industry evolve and we keep on building towards the future. You can expect some launches soon, but I want to start introducing how we all got here and where the industry is going. The vocabulary changes faster than the value The industry has been cycling through terminology. Frameworks became scaffolds. Scaffolds became harnesses. The idea is that each generation is more opinionated. Harnesseses ship with planning, memory, file systems, and compaction built in. But the underlying dynamic hasn't changed and meanwhile the building layer keeps getting cheaper. @garrytan put it well this last week: the harness should be thin. The harness is plumbing. Plumbing matters, but nobody builds an exciting defensible product on plumbing. Meanwhile, some are arguing that harnesses are the new defensible layer, that frameworks are dead but harnesses will endure. I think that's wishful thinking. The same forces that commoditized frameworks are already starting to commoditize harnesses. Model providers keep absorbing the stack and every quarter, another primitive moves behind an API. We have a framework ourselves (CrewAI Flows) and a Harness (CrewAI Crews and Agents, with memory, tools, caching, context engineering, prompts, skills, mcp, etc.. you get it), so I'm not saying this from the sidelines, I'm on the front row watching it happen to a category I'm helping build. The entire conversation is furniture rearrangement. So if the harness is dying, what lives? The answer isn't another layer of tooling. It's something else entirely. The building is getting cheaper and that affects incumbent SaaS as much as it affects startups and enterprises, rearranging the furniture doesn't change that. The things that compound When building gets cheap, the value moves to the layers that can't be replicated overnight. Distribution. Proprietary data that took years to assemble. Products that capture intelligence as customers use them and get better over time. Trust earned through production. It’s usually a combination of these things that create separation between great and mediocre software nowadays.…