Who Owns the Why

There's a simple test for whether an AI agent is actually useful or just impressive in a demo: give it a task that requires knowing why your team does things a certain way.

Most of them fall apart immediately.

Not because they can't execute. They can execute fine. They fail because they have the what (tickets, code, metrics) but not the why. And the what and the why are deeply coupled. You can't separate them and still have something useful.

A task management agent that can see every open ticket but doesn't know you deprioritized that entire product area six months ago for reasons that aren't written down anywhere. That agent is going to make confident, wrong suggestions all day. The same goes for any agent operating inside an organization with any real history. The why is invisible to it. So it keeps bumping into decisions it can't explain.

That's the problem nobody's really talking about. We're spending enormous energy making agents more capable. But capability without context is just a faster way to go in the wrong direction.

The Toll Road

Here's what I think happens over the next few years. Context becomes the most strategically important asset in software. Specifically the why behind how an organization operates.

Not the model. Not the interface. The context.

Without proprietary context, every software product is a thin wrapper. A nice UI on top of whatever data you feed into it. Swap out the wrapper and the data stays. So does the real value. The product that owns the context owns the toll road. Every agent, every workflow, every team member has to pass through it to do anything useful.

Think about Salesforce. The reason it's hard to displace isn't the features. It's that every piece of customer context lives there. Your pipeline history, your deal notes, the reasons you lost that enterprise account three years ago. Any AI tool that wants to be useful for your sales team has to either integrate with Salesforce or accept that it's operating blind. That's the toll road.

Notion is positioning itself similarly for internal knowledge. The org that has years of decisions, context, and reasoning living in a well-maintained workspace is going to get dramatically more value from AI agents than the org whose institutional knowledge is scattered across Slack threads, email chains, and the heads of people who've since left.

What We Lost When We Flattened Orgs

Here's the thing about middle management that the flat org movement missed: not all managers are created equal.

The bad ones were real overhead: approval layers, status meetings, people who moved slowly and added friction without adding value. Getting rid of them made sense.

But the great ones were multipliers. They held the why. They knew why the API was designed that way, why that customer segment was deprioritized, why the team made the architectural call they did three years ago. They translated context between teams, kept people building in the same direction, and made fast execution safe because everyone understood what they were optimizing for.

When orgs flattened, both went. The overhead and the connective tissue.

What a great manager was really doing, underneath all the meetings and reviews, was providing context and direction. And context is direction. When people understand the why clearly enough, they don't need to be told what to do next. The context orients them. The manager's job was to make the why so legible that direction emerged naturally.

We removed that layer and bet that culture and documentation would fill the gap. They mostly didn't. Documentation gets stale. Culture is implicit. The why started leaking out of organizations quietly, and most teams didn't notice because execution could still move fast enough to paper over the gaps.

AI Makes the Problem Acute

AI is about to make this much worse before anyone fixes it.

Production is flattening. Output is going up: more code, more decisions, more features, faster. The connective tissue problem gets exposed when output accelerates. You can move fast with fragmented context for a while. At some point the pieces stop fitting together.

Without managed context, you get disjointed software. Teams building things that don't contradict each other on any single ticket but don't add up to a coherent system. Agents making locally correct decisions that are globally wrong because each one is operating against a different slice of organizational reality.

The coherence problem isn't new. But AI makes it urgent. The volume of decisions is going up. The cost of each individual decision is going down. But the cost of incoherence, of all those cheap fast decisions pointing in slightly different directions, compounds faster than ever.

The Internal Wiki Is About to Have Its Moment

This is why I think internal wikis are going to go from that thing nobody maintains to genuinely fought-over infrastructure.

Right now, most internal knowledge bases are a graveyard. Pages that were accurate in 2021. Onboarding docs nobody updates. Architecture decisions buried in Confluence somewhere that the people who wrote them have long forgotten. The why is technically written down, but it's not structured, not maintained, and not useful to a machine.

That's going to change. The organizations that treat their internal knowledge base like a database (structured, maintained, machine-readable) are going to have a real advantage. Their agents will know why things are done a certain way. Their new hires will get up to speed faster. Their decisions will compound instead of fragment.

The internal wiki becomes the context layer. The toll road. The thing every agent passes through to understand what direction means in this specific organization. It's what a great manager used to carry in their head. Now it needs to live somewhere that scales.

Fewer Foot Soldiers, Higher Stakes

AI is eliminating a certain kind of developer, not because they're not skilled, but because what they do is execution. Writing code to spec. Following a ticket. Implementing a defined interface. That's the robotic part of development, and it's exactly what agents handle best.

Engineering culture has historically rewarded those traits. Ships fast. High output. Doesn't slow things down asking too many questions. Those are the people who got hired and promoted. They're also the most exposed right now.

What becomes irreplaceable is the judgment: the ability to hold the why and ask whether the spec itself is right. The developer who pushes back on a ticket because they understand the broader context. The one who notices three features are solving the same underlying problem. The one whose understanding of the why keeps everyone oriented.

This matters more as teams shrink. When AI handles execution and you need five engineers instead of twenty, misalignment is catastrophic. Five people building in slightly different directions, each amplified by AI, fragment faster than a large team ever could. Shared vision isn't a nice-to-have at that scale. It's the actual bottleneck.

Context is what creates that alignment. When the why is legible and distributed, whether through a great manager or a well-maintained knowledge base, a small team can move fast and stay coherent. Without it, speed becomes the enemy. You're just fracturing faster.

The developers worth betting on aren't the most prolific. They're the ones who understand what they're building toward and can make that clear enough for everyone else, human or agent, to orient around it.

Who Owns the Why, Owns the Flow

The value question for any software product right now is simple: does this product capture the why?

Not just the what. The what is already well-served: ticketing systems, CRMs, analytics tools, databases full of what happened. The gap is the why. Why that decision was made, what constraint was operating at the time, what the team was optimizing for, what they explicitly chose not to do and why.

Whoever captures that, in a form that's structured and accessible to agents, owns something genuinely hard to replicate. You can copy features. You can't copy years of organizational reasoning.

The shift isn't about AI capability. It's about context infrastructure. The teams that build it, or the tools that help them build it, are the toll roads of the agent economy. Everything useful passes through them.