Rob Yeldham (he/him)

Something very important for most people to understand is now happening inside parts of the tech industry right now. Companies aren’t just automating work. They’re systematically extracting it. Across firms like Atlassian, Amazon, and Block, a new pattern is emerging: Knowledge extraction sprints. These are structured, time-bound efforts to capture how top performers actually do their jobs in the wild. This is not sanitized documentation, viability testing, or training decks, but real, proactive cross-training of AI systems on what workers actually do. And then using AI agents to do the work humans previously did instead. Not documentation. Not playbooks. Action. Doing the work. • Screen recordings are made of live work • Decision paths and tradeoffs are captured • Debugging flows and problem-solving sequences saved • Even strategic judgments in real time This is a key missing ingredient for agentic AI. The constraint is no longer model capability, it’s access to how work actually gets done inside high-performing teams. These extraction efforts convert human expertise into machine-readable workflows that agents can execute. That’s the real shift and it has major pros and cons. We have started moving from tools that assist work… to systems that precisely replicate how work is done. Once captured, that knowledge doesn’t just scale. It detaches from individuals, from teams, and eventually from labor itself. The organization begins to move from human-executed to system-executed work. This is where the slope gets quite slippery. Because the message to employees is increasingly clear: Your job is no longer just to perform work. It is now to teach the system how to perform the work instead. That is a very different psychological contract, and it breaks trust much faster than most leaders realize or anticipate. We must now expect rapid changes in employment norms: • Fewer senior roles anchored in experience • More junior + agent combinations executing work • Institutional knowledge shifting from people into systems For CIOs, this is not about AI adoption. It’s about who owns execution inside the enterprise. If vendors or external platforms are capturing and encoding your workflows, you are effectively outsourcing your operating model. Governance alone will not solve this. You need new operating structures that are designed for this reality: • Human and agent collaboration as a first-class design principle • Internal ownership of critical workflows and decision logic • Agent orchestration as a core enterprise capability This is organizational architecture now, not tooling strategy. For technology leaders, the tradeoff is even sharper. Knowledge extraction at scale can create enormous leverage — but it can also destroy trust and actually lose institutional knowledge if handled poorly. And without trust, the flow of high-value knowledge work will slow or stop entirely. The enterprises that navigate this successfully will be explicit about the new contract: • What is being captured and why • What remains uniquely human • How employees evolve and gain leverage in an agentic system • Basic fairness on how workers are treated Because the end state is no longer theoretical. • Agents execute • Humans orchestrate • Organizations become programmable And right now, we are watching the first companies figure out how to capture, and ultimately own, how work actually happens. We should learn how to do this right, while ensuring humans still have a strong stake in the resulting systems.