Matt Burgess

I set out to build an instrument to help customers be heard, and to help teams be better at their jobs. Then I realised that the instrument I’d built was a thing: a boundary object.

I came across the term through the work of Dr Jabe Bloom. It comes from a 1989 paper by Susan Leigh Star and James Griesemer, who studied how a natural history museum actually works. They looked at how a single specimen - a pinned butterfly, say, or a fossil - does different work for different people inside a museum. The person who collected the specimen in the field engaged with it one way. The taxonomist, placing it against a wider classification system, treated it another. The curator, responsible for storing and preserving it, handled it a third way. The educator, explaining it to a visitor who’s never given insects or geology a moment’s thought, called on it a fourth way. Star and Griesemer called this a boundary object: something adaptable enough that each community can shape it to their own use, while it stays robust enough at the centre to remain recognisably the same thing to all of them. What that buys everyone is easy to miss and hard to overstate: nobody has to wait for anyone else to agree before they can act on what they see.

Here’s another example: a cricket pitch is a boundary object. And bear with because you don’t need to know anything about cricket for that to make sense.

A cricket pitch wears down over the course of a match - the same strip of ground, scuffed by hundreds of footfalls and however many deliveries have landed on it by the second day. The bowler sees an opening in the worn patch and bowls into it, right now, at this particular point in the day. The batter reads the same patch and adjusts their stance just as fast. The groundsman reads the scuff marks as a forecast for how things might play out tomorrow. The umpire logs it for a ruling that might not come until much later in the match, if it comes at all. Unlike the museum specimen, which is already fixed by the time anyone gets to it, the pitch is still forming while it’s being read - by people acting on their own timing, none of them needing the others' permission first.

Ask a large organisation to point at something like this in its own world - something everyone can act on without waiting for the others to agree - and it will usually point at a slide deck. A research team runs a project, produces a deck, and the rest of the business - product, marketing, loyalty, whoever needs it - is meant to read it and work from it. On the surface, that looks like exactly what I’ve just described.

But a deck isn’t a boundary object. It’s a boundary object-shaped object, and it fails on three things that actually make a boundary object work.

The first problem: every function reading a deck is reading somebody’s synthesis of the original material, not the material itself. The product manager doesn’t watch a customer feedback interview. She reads a slide about a summary of the interview. That isn’t four communities finding their own thing in a shared source - it’s four communities reading the same secondhand account and hoping it happens to answer their particular question. Something changes shape at that handoff, the way it does at every handoff. It’s called transduction. A deck’s job is to present, not to hold - so it summarises away plenty that might be exactly what another team needs, or what even this team needs in a month’s time. What survives the crossing into and out of a deck is rarely what turns out to be useful to somebody else, someday.

The second: take the synthesis out of a deck and there’s nothing underneath it. What was actually said was never shared in the first place - only somebody’s account of it was. Remove the second or third hand account and there’s nothing left to fall back on. That gap has a real cost, and it’s one I’ve written about elsewhere as signal debt - it behaves like debt in the ordinary sense, building up quietly at every point the signal changes hands, and nobody notices the size of it until much later, if at all.

The third: what actually gets read out from a deck was never one person’s expression to begin with. Research obtains good stuff from a range of people, and a deck and every other subsequent derivation presents something that sounds like a single customer talking - smoothed, averaged, built to stand in for everyone rather than to be anyone in particular. There’s no actual individual behind it for any of those four communities to be reading, even in the version where the material is shared honestly. What I’ve been calling canonical customers elsewhere is the alternative: each actual, individual thing a real person expressed, treated as its own authority, rather than blended into a composite built to represent many people at once.

Because here’s the thing: AI doesn’t fix any of this. It makes all three worse: more transduction loss, more signal debt, and really rapid contamination by AI slop and bots, which then get summarised and round we go again. All of which leads to the conclusion that a shift to working from genuine boundary objects is more urgent than it’s ever been.

What a deck can’t hold - because holding was never its job - is worth holding onto. Every action and decision needs provenance, and organisations are making more of those decisions, faster, with AI doing the summarising now too. Customers deserve to be heard, and teams are better when they hear them properly - and better, for teams, means their own read at their own speed, not anyone else’s. Stratum is what I built to do that: it holds what a customer actually expressed - not a summary of it, not somebody’s read on it - so whoever needs it can go straight to the thing itself. Building from the source keeps decisions customer coupled - tethered to what was actually said, not to whichever version of it has drifted furthest away. It turned out to be a genuine boundary object. I’m happy with that.

Follow along: mattburgess.micro.blog/subscribe… mattburgess.micro.blog/feed.xml micro.blog/mattburge… Mastodon @mattburgess@micro.blog