There is a huge opportunity for resourceful and entrepreneurial talent within organizations to go in and reimagine workflows for a world of AI agents. The way you automate work with agents requires real work. It means setting up unstructured data in a way agents can easily access, learning the workflow and processes and creating skills or plans for agents to leverage, connecting disparate systems together, and likely changing the process itself to support getting the agents the need to do much of the work. Then you have to design where humans will play a role to oversee the workflows, how you validate the work, and so on. Most of the gains you see from coding don’t take this level of effort because the agent knows more, it gets context more easily, and the users are technically. But for the rest of knowledge work there’s no way around this; there’s really no way to shortcut any of this work. It has to be done by a person or people on the team. You will see a huge growth of roles within enterprises, and people that specialize in this will be hugely valuable in the economy. Great way for early career folks to make a huge dent quickly as well.
Agents usually expose the same constraint we see in operational systems — outcomes depend less on the technology and more on how workflows and decision paths are structured. Predictability is designed into the system.
This is the part most orgs are sleeping on. The people who figure out how to redesign workflows around agents will create way more value than those just plugging AI into existing processes.
I had this epiphany in march of ‘25 Over a decade spent building products Products began to drift to wrap models Products use to be the wrapper for use-cases The complexities compressed My edge? I was building products faster than any one could believe but if I can do that? then can’t everyone in 12 months? ofc products started to wear a hat loaded with so much more risk Product traction windows got smaller = risk Product pricing shrunk = risk There were products out there that had raised millions to be just that: a product yikes Question i asked: if product cycles compressed then where is the shortage? It won’t be in products. It won’t be in founders behind products. It won’t be in opex costs with agentic development It will be in entrepreneurs needed to function within a organization so that the org can truly become agentic There are levels to this: 1. Using AI 2. Using agents 3. Governing agents 4. Orchestrating agents 5. Repeatable systems of agents 6. Observability It’s going to be the ex-founders haven The exhilaration. The shared risk. The upside will look different. Happy to be here for the ride
This is solving for capability, not control. You can structure data, encode workflows, and build all the skills you want. That just makes agents better at producing answers. It doesn’t govern what they actually do. The failure shows up at execution. Agents drift, over-agree, or contradict each other, and the system still produces a single action with no real notion of admissibility. Human validation doesn’t fix this. It just audits the aftermath. The next phase isn’t more workflow design. It’s a control layer at the execution boundary that decides what is allowed to happen before it happens. Without that, we’re scaling intelligent systems without a mechanism for accountability.
Hard agree. I am this persona. It’s a lot of work — and fun to boot.
In most environments, over half of operational data is unstructured or poorly standardized. Until that is addressed, agent performance will plateau regardless of model quality, which is why the bottleneck is shifting from technology to implementation.
Very well put. For many organisations, this post captures the number one job to be done - “work on the work”.
This feels like a new kind of operator role: part engineer, part PM, part systems thinker. The tricky part is that you can’t abstract this work away; you need deep context of how the business actually runs. That’s why it’s such a high-leverage opportunity right now.
100% For a client our team built a modern data stack from scratch with fully automated orchestrated data pipelines & data schemas that is optimized for downstream agentic applications. Then we build the agentic application(s) on top of the data foundation. The data layer needs to be rebuilt as well. We help teams build AI-first data systems & agentic solutions. If you're a growth stage team or business exploring how to leverage agentic workflows, message us to learn how we can build data systems & agentic workflows for you. #agentic LangGraph AI Anthropic
Aaron Levie Completely agree, there’s a real shift happening in how workflows are designed, and it’s much more involved than just plugging in a model. At the same time, as agents get access to more systems, data, and decision-making, the risk surface grows just as quickly. It’s easy to focus on capability and miss how these workflows behave in real scenarios until something breaks. Feels like designing these systems now needs to balance both execution and control from the start.