Where Should AI Work Actually Happen?

On a call last week, a partner asked me, slightly sheepishly, whether I still keep my to-dos on pen and paper. I do. When something needs actual thought, writing it down is the thinking.

It was one version of a question we field on nearly every DealSage call: where should AI work actually happen? Should the memo be drafted in the platform or in a chat window? Should the model live in Excel or behind an agent? Should meeting notes be queried where they were taken, or somewhere central?

The honest answer is that it depends, and that the question is aimed at the wrong layer. Where you do the work matters far less than where the work ends up.

The tools are front doors

Here is a plain inventory of where my own work started in a single recent week. A notebook. Voice notes dictated in the car. Email. A chat assistant when I wanted to think something through, a coding agent when I was building, agents inside our own platform running on their own schedule. Some things simply feel right in a particular tool, and the feeling shifts with the task.

That flexibility is worth leaning into rather than fighting. Nobody should be hunting for the one true app, because the honest experience of working with AI in 2026 is that different surfaces are good at different things, and the gaps move every quarter.

The reason the flexibility works, though, has nothing to do with the tools. Every one of those doors pulls from the same underlying record. A session never starts with re-explaining who we are, what the deal is, or what was agreed last week, because the context is already there whichever door you walk through. And everything that matters lands back in the same place: tasks routed to whoever owns them, ideas into a database, deal work onto the deal record.

The tools are front doors, and front doors are disposable. Swap any of them tomorrow and nothing downstream notices. The record underneath is the part you can’t swap, and it’s the part most firms haven’t built. We’ve written before about why AI needs a system of record, and everything in this piece rests on that foundation.

Skills make the surface irrelevant

There’s a second thing that makes multi-tool work viable, and it took longer to see. When agents produce most of the output, the thing a firm actually needs to standardise is how it likes things done: how a memo should read, what a model must always include, which numbers get checked and against what.

This now has a name, skills, and it’s fast becoming common vernacular across the AI tools. A skill is your way of doing something written down, with examples to pull from, in a form any agent can follow.

Once that exists, it stops mattering which surface produces the work. The same skill yields the same memo whether it ran from a chat window, an agent inside the platform, or a scheduled job overnight. Consistency used to live in a senior person’s head and a house-style deck nobody opened. Now it travels with the work.

The two ways firms get this wrong

The first failure mode is work that evaporates. Drop a data pack into a chat model and you get a decent spreadsheet, and six weeks later nobody can trace a single number in it back to a source document, because the work lived and died inside a thread. A transcript is a record of a conversation. It is not a record of the work.

The second failure is the opposite: a platform that insists everyone comes to it. Sell-side teams that move buyer Q&A into a portal watch outside advisors refuse to log in, and the whole process reverts to Excel over email within a fortnight. Work has to be caught where it already happens, in the inbox, in the spreadsheet, in the meeting, because change management loses to habit almost every time.

Both failures are the same mistake seen from opposite ends. The tool was judged as a destination, a place people must go, rather than as plumbing that catches work wherever it starts.

This is what a working landing zone looks like in practice. One PE client’s partners can ask “why did we pass on this deal in 2021?” and get an answer in seconds, because years of deal history, memos and decisions were structured into one queryable record. The question can arrive through any door, chat, email, an agent, and the answer comes from the same place, with lineage back to the documents behind it.

The test to run before you buy anything

Start a piece of work in your inbox. Continue it in a chat window. Finish it wherever your team would naturally finish it. Then ask two questions.

First, how many times did you re-explain something your firm already knows? Every re-explanation is a tax you’re paying for a record that doesn’t exist yet, and it gets levied on every person, every session, forever, until the record gets built.

Second, where did the work end up? If the answer is “in the chat window”, you have a transcript, and transcripts don’t compound. If the answer is a structured record your whole team can query, with every figure traceable to its source, then the tools on top can change as often as the market changes them, and it costs you nothing.

The nature of work keeps evolving: typed to spoken, assembled to generated, opened-in-an-app to delivered-by-an-agent. Betting on any particular channel is a losing game. Betting on the record underneath them is the same bet every time.

I still like the notebook, for what it’s worth. It’s the one front door that never gets a software update.