AI as context holder: The productivity gain nobody predicted

AI did not make the work easier. It made remembering easier. And that turned out to matter more than anyone expected.

For all the discourse about AI generating content, writing code, and producing strategy documents, the most useful thing AI has done for my productivity this year has nothing to do with generation. It holds context. It remembers where I left off three weeks ago, what decisions were made, what open questions remained. That is not glamorous. But it has changed how I work more than any feature demonstrated on a conference stage.

The conversation about AI and productivity has been fixated on output. How many words per minute. How many lines of code. How many slides. But the bottleneck for most product people was never output. It was the cost of returning to a problem after being away from it.

Once upon a time

For most of my career, I managed between three and five parallel workstreams at any given time. This is normal for product people. You have the main initiative, the exploratory track, the partnership discussion, the team issue that needs quiet attention, and the thing your VP asked you to look into that is not quite a project but is definitely not nothing. Each lives in a different context: different stakeholders, different documents, different stage of thinking.

The cost of switching between them was never the switching itself.

It was the re-entry.

I call this the re-entry tax. Every time you return to a workstream after time away, you spend the first twenty to forty minutes reconstructing the mental state you had when you last worked on it. Where were we? What had we decided? What was I about to investigate before I got interrupted? The answers are scattered across Slack threads, meeting notes, documents with eighteen comment threads, and your own memory, which is the least reliable of the four.

At Grab, I was running two product tracks while managing a cross-functional dependency that required weekly negotiation with three teams. Every Monday morning, I sat down intending to pick up where I left off on Friday. But Friday's context was gone. Not the facts. Those were in documents somewhere. The thinking. The particular way I had been framing a problem, the angle I was about to explore, the intuition I was developing about a user behaviour pattern. That kind of context does not survive a weekend. It barely survives a lunch break.

I once spent an entire Monday morning re-reading my own notes from the previous Thursday, trying to reconstruct a line of reasoning that had felt clear forty-eight hours earlier. By noon, I had recovered roughly seventy per cent of it. The other thirty per cent was gone. Not lost in a dramatic way. Just quietly evaporated, like a conversation you mostly remember but cannot quite quote.

That thirty per cent, across dozens of workstreams over months, adds up to a staggering amount of lost cognitive work.

Every day, until one day

The standard solutions to context-switching cost were all organisational. Block your calendar. Batch similar work. Reduce your parallel tracks. Good advice, genuinely. But impractical for most product roles, where the parallel workstreams are not optional. You do not run five tracks because you enjoy it. You run them because the work requires it.

Earlier this year, I started doing something different. Instead of closing a work session and trusting my future self to remember the state of play, I began recording the session's context in an AI tool. Not a summary. A state capture. Here is where we are. Here are the open questions. Here is what I was thinking about exploring next. Here is the decision I was leaning toward and why.

The first time I returned to a project after a week away and interrogated the context capture, the difference was immediate. Instead of forty minutes of re-entry, I was back in the problem within five. Not because the AI had solved anything. But because it had held the context that my brain had discarded. The thinking was preserved in a form I could re-enter, rather than reconstructed from fragments.

I had accidentally built what I now think of as a persistent context architecture: a system where the cognitive state of a project is captured and made retrievable between sessions.

Simple in design. Disproportionate in impact.

Because of that

The implications became clear within weeks. With the re-entry tax reduced, I could switch between workstreams without the cognitive penalty that had previously made parallel work so draining. My work on secondary tracks improved because I was no longer starting each session partially amnesiac. But more importantly, I started noticing things across workstreams that I had missed before. Patterns. Connections. A user insight from one project that was relevant to another. Those connections had always existed. But the re-entry tax had consumed the bandwidth that would have spotted them.

At Schneider Electric, years earlier, I had watched a senior product leader maintain what she called her "project brain" in a paper notebook. Every Friday, she spent thirty minutes writing a status update to herself. Not to her manager. To her Monday-morning self. Here is my current hypothesis. Here is what I plan to test. Here is the thing I keep avoiding thinking about. It was brilliant. But it was manual, time-consuming, and dependent on her discipline to maintain it every week.

AI makes her method scalable.

The same practice, without the friction.

Until finally

The pattern I am seeing across PM communities is not people using AI to generate product briefs or write user stories. It is people using AI to hold project context between sessions. The use case is unglamorous. It will never make a keynote demo. But it addresses a specific, measurable, daily problem that every product person who manages parallel work recognises the moment you name it.

The re-entry tax is real. We have all paid it. But most of us pay it multiple times a day without recognising it as a cost, because it disguises itself as "getting back up to speed," as though re-establishing your mental state were a natural part of the work rather than pure overhead.

What has changed is that the tax has become optional. Not eliminated. But dramatically reduced for anyone willing to spend three minutes at the end of a session capturing where their thinking stands. The value is not in AI doing the thinking. It is in AI holding the context so you can think faster.

Ever since then

I have been mentoring junior PMs for years, and the advice I used to give about context-switching was defensive: reduce your parallel tracks, protect your focus time, batch similar work. Good tactics. But they treat the symptom without addressing the cause. Human memory is not designed to hold the cognitive state of five parallel projects across a week. We forget. Not the facts. The framing. The angle. The almost-formed thought.

AI as context holder is not a breakthrough in artificial intelligence. It is a workaround for a limitation in biological intelligence. And workarounds, in my experience, tend to be the innovations that change daily practice. Not the spectacular ones. The practical ones.

The most important thing AI has done for my work this year is something I would have found absurd to predict twelve months ago. It did not make me faster at writing or better at analysis. It remembers where I was. And that is what made the difference between five workstreams that drained me and five workstreams that fed each other.

Some tools change what you can do. The best ones change what you can hold in your head while you do it.