As AI agents become a core part of enterprise workflows, their success hinges on one thing: context. Unlike humans, agents can’t rely on ambient knowledge or informal handoffs. In this post, Bodhi Venture Labs founder Angus Norton draws from his experience at Microsoft and Amazon to explore how product and engineering teams can operationalize context — through better documentation, structured data, and system design — to fully realize AI’s potential today.
In my last post, I explored a fundamental shift: AI agents are only as effective as the context we give them. These systems don’t inherit tribal knowledge. They don’t sit in meetings. They don’t absorb culture or infer strategy unless it’s made explicit.
If you're in product or engineering leadership, this has real implications. The way we build software, manage roadmaps, document processes, and even define success—all of these aspects need to evolve to support agent productivity. It's not about retrofitting AI into what you already do. It's about retooling how your teams operate so AI agents can actually contribute.
This is where operationalizing context becomes your strategic advantage.
We've long accepted that enterprise documentation drifts, onboarding is challenging, and that most organizational knowledge resides in people's heads. But in a world of agents, that tolerance becomes a liability. If an agent can’t “look over someone’s shoulder” or “ask around,” then the only way it can be effective is if your environment makes context explicit, structured, and retrievable.
This isn’t a philosophical shift—it’s a practical one. Companies that want to harness agents today (not in five years when things “just work”) will need to intentionally design for this constraint.
And it starts inside your product and engineering organizations.
In my experience with large and small companies over the last 30+ years in this business, I have learned that at the heart of this transformation is a mindset change: you’re no longer just building for users and developers—you’re building for agents too.
That means the specifications you write, the workflows you model, and the systems you architect all need to be legible to machines. Not just comprehensible in a loose sense, but structured enough to be usable without human interpretation. Amazon's working backward mechanisms are world-class in this regard, and they have shaped my approach since I worked there as a product leader.
In the same way that clean code benefits future developers, clean and structured context benefits agents. Every time your team writes a well-formed user story, updates a test case in a standardized format, or defines a workflow with clear decision points, they’re not just helping future teammates—they’re enabling automation to work.
In many ways, documentation is becoming the new runtime environment for agents.
Just like code, it needs to be versioned, maintained, and regularly reviewed. The days of dumping tribal knowledge into a wiki and calling it “done” are over. Living documentation—tracked in your source control, updated alongside your product, and written in a way that machines can parse—is becoming critical infrastructure.
This means embedding documentation into your development flow, not treating it as an afterthought. It also means thinking beyond prose. The most useful documentation for agents is semi-structured or structured—think JSON schemas, API specs, Markdown with front matter, or YAML-configured workflows.
What’s changing here is not just the tools, but the role of documentation in the development process. It’s no longer just for onboarding humans—it's the instruction manual for your AI workforce.
Context isn't limited to what you write down. It’s also about the systems of access you design.
Imagine an agent who needs to complete a task, such as generating a report, automating a customer onboarding flow, or summarizing internal feature requests. If that agent can’t reliably locate the right dataset, understand what the data means, and know what action to take with it, it’s dead on arrival.
So, how does your product handle metadata, access controls, versioning, and data modeling? It matters. A lot. We’re moving toward a world where your data architecture is as much about clarity and retrievability as it is about storage and performance.
The companies winning here are treating context access the way they treated performance optimization ten years ago: as a core product capability.
All of this amounts to a fundamental evolution in product management.
Historically, we’ve defined our job as shipping features that solve user problems. That’s still true—but now we’re also designing for collaboration with non-human teammates. In this case, AI agents.
This doesn’t mean you need to become an AI expert. But it does mean you need to be fluent in how context works in an AI-native world.
For example:
You’re no longer just building interfaces. You’re building systems that must be navigable by both humans and machines. That requires clarity, consistency, and intentionality at every layer.
Currently, most companies are experimenting with AI through a prompt-centric approach. “Let’s see what ChatGPT can do with this.” But the organizations that are truly advancing are taking a system-centric approach.
They’re building internal tooling, workflows, and knowledge graphs that enable agents to operate in a defined and repeatable manner.
That doesn’t mean every system has to be rebuilt from scratch. It means evolving your current systems to expose more of your internal logic and workflows in a structured, machine-usable format.
And it means training your team to think less like “human collaborators” and more like orchestrators of intelligent systems.
In the long run, AI agents won’t just be helpful assistants—they’ll become the primary interface through which your business logic, operations, and knowledge are executed.
Just as APIs have become the standard way to access software services, agents will become the standard way to activate business functions. But they can only do that if they can understand how things work. Which is why context design is the new competitive advantage.
Personally, this is the culmination of a journey I’ve been on for years—starting as a product manager on Microsoft SharePoint, where we tackled the earliest problems of knowledge management. Then, through Office 365, we reimagined cloud productivity. Then to Bing, where we learned how to connect people to the proper knowledge at scale. And later at Amazon WorkDocs, where we worked on turning static content into active collaboration.
In every chapter, I was chasing the vision Bill Gates spoke about decades ago:
“Information at your fingertips.”
That dream is no longer aspirational. With AI, we are finally getting there. But only if we build systems that make information understandable—and actionable—for machines.
At Bodhi Venture Labs, I help software and services companies re-engineer their product and GTM strategies for an AI-native world. If you’re ready to go beyond the hype and start operationalizing context in your product org, let’s talk.
Comments settings
Like
Comment
Share
