· Charlie Holland · Architecture · 8 min read What a company really spends its money on is the gathering: getting everyone pointed the same way, getting a decision to actually happen, getting one change past four teams and out of the door. More work has always meant more people, and more people has always meant more gathering. A working sheepdog doesn't obey that arithmetic. One dog and one handler move several hundred sheep, and another thousand sheep doesn't mean another thousand dogs. That ratio is the thing AI could finally break, and almost nobody is building for it yet.
Half past six in the morning, and The Fonz reached the window before I reached the door. He’d clocked, a clear second ahead of me, half a dozen sheep on the wrong side of a fence. Then all merry hell broke loose.
They aren’t my sheep. I’m not a farmer, I just live in the middle of farmland, and every so often half a dozen of next door’s flock find the gap, the way they always find the gap. This morning they’d taken it early and were already ambling downhill with the unhurried confidence of animals who’ve decided exactly where they’re going: the road, five hundred metres away, the one direction that matters because it’s the one with cars on it.
So there I was, giving chase across a wet field in Crocs and a dressing gown. A sight, I’m sure, for anyone unlucky enough to be watching.
I should be honest about the dog. The Fonz is a Briard, a French sheepdog, and he looks the part completely. He’s also ten years old and has, in his entire distinguished career, herded precisely nothing. He’s a chaser. Let him loose at six sheep and he’ll pursue them with tremendous commitment in the exact direction they were already going, which this morning was the road. Enormous enthusiasm, no plan.
Drying off afterwards, I noticed something. Not one part of that was a speed problem.
Chasing isn’t herding #
It wasn’t a speed problem because a sheep, over five hundred metres, is faster and far more motivated than a middle-aged man in Crocs, and always will be. A quicker dog wouldn’t have helped either. It just reaches the wrong place sooner and panics the flock when it gets there. The Fonz brought plenty of speed. Speed was never what the morning was short of.
What the morning needed was the thing I don’t have: a dog that herds. A working collie doesn’t outrun the sheep, it reads them, and puts itself in the one spot that turns all six back through the gate before they reach the tarmac. A different job altogether.
That job, the gathering, is what a company spends most of its money on. Getting everyone pointed the same way. Getting a decision to actually happen. Getting one change past four teams and out of the door. It does show up on the plan. We call them milestones, which is a polite word for a date by which everyone is supposed to have finished agreeing. And all of it has to be done by somebody.
I’ve written two of these already about where the money really goes. Complexity is never removed, only relocated, and a growing share of every gain breathes away inside the organisation before it reaches anyone. Both land in the same place: it was never how fast any one person worked, it was the cost of holding the whole thing together. And that work has only ever had one tool. People. Every person you added to help hold it together was one more thing to hold together.
Everyone’s building a faster chaser #
So what are we all building? Chasers.
And not just the vendors. Almost every developer I know is building their own: harnesses, wrappers, custom agents, one more slash command to shave thirty seconds off a thing they do four times a day. Open X or Reddit on any given morning and half the feed is somebody’s new rig for going faster.
They work, too. The coding assistant, the chatbot, the thing that drafts your emails and summarises the meeting you didn’t want to sit in. A good one makes a developer feel twice as fast. But every one of them is pointed at the same spot: one person, doing their own piece of the work, a bit quicker. That’s The Fonz. Enormous commitment, genuine speed, applied in whatever direction things were already travelling.
And a keener chaser scatters the flock harder. A faster developer ships more code, which is more to review, more to integrate, more to release, more to square with the four other teams now shipping faster too. A quicker analyst turns out more analysis for more people to read and reconcile. All of it lands on the same people who couldn’t keep up with the old volume. We’re kicking the can down the road and hoping a technological miracle turns up to save us.
None of which is an argument against the tools. We’re just buying the wrong half of the problem.
One dog, three hundred sheep #
Everyone expects AI to make us more productive, and everyone is right. There’ll be more work, not less. The work always grows, every technology we’ve had raised what a business could attempt and then watched it attempt exactly that much more, and nobody gets a machine that does half their job and goes home early. The flock gets bigger.
The question is how much bigger, and that’s set by the gathering. In every organisation I’ve worked in, the only way to hold more together has been to add people, which adds more places for the work to fall between them, which adds more gathering. The ratio has never moved.
A working collie doesn’t obey that arithmetic. One dog and one handler move several hundred sheep across open ground, because the dog is doing geometry rather than sprinting, and another thousand sheep doesn’t mean another thousand dogs. More work held together by the same few people: that’s the prize everyone thinks they’re buying with the chasers, and it’s sitting untouched.
We’re not there yet #
Nobody has this yet. What you see today, when a team gets proper AI assistance, is the chasing getting faster. Give a few good developers real help and they start freely changing code and infrastructure they’d never have gone near before. Put three or four of them on one product and you get constant churn, one developer’s agent quietly reworking what another’s added the day before. The agents do exactly what they’re asked, and what they’re asked depends entirely on who’s doing the asking.
Six chasers on one flock, every one of them keen, no two of them working to the same plan. The sheep end up in every corner of the field.
The gathering side has barely started. A handful of outfits are building for it directly now rather than selling another assistant, and they’re early. Every technology spends a while like this, working brilliantly at the thing it was first pointed at while the shape of the work catches up behind it.
Move your people to the gathering #
So what do you actually do, if you’re running something?
The instinct is to buy every chaser going and hand one to everybody. Accenture and Anthropic announced exactly that last December: thirty thousand Accenture people trained on Claude, tens of thousands of them on Claude Code, which Anthropic called its largest deployment to date. Thirty thousand faster chasers. It’ll work, too, in the sense that thirty thousand people will each get quicker at their own piece of the job, and every one of them will be putting more work into the same field.
The move is to get your people out of the chasing and into the gathering. The chasing is the part that’s being automated, and it’s being automated well. Doing your own individual piece of work faster is what every tool on the market already does, and it does it better every quarter. What holds its value is the thing the collie does: reading where the whole thing is heading and knowing where it ought to go instead. Which way to turn it. When to apply pressure and when to stand still and let the mass carry.
And you’ll need fewer of them. That’s what a ratio means: a few dogs and a very large flock. The people whose real job was holding the complexity together, the ones the last post called the angels, end up needed in far smaller numbers and worth a great deal more.
I’ve spent years being the person a business brings in to hold a complicated thing together, to work out where it’s all heading and get it pointed the same way. That’s gathering. So when I say the gathering is the part worth automating, I’m arguing for the thing that puts me out of work, and I’d rather you heard that from me than worked it out afterwards.
It’s still worth doing. The flock is getting bigger whatever anyone decides, and a flock that size with nothing but chasers on it ends up on the road.
Which leaves one question, and it’s where I’ll finish the series. If the dogs turn out to be something anybody can buy, and in this industry everything eventually is, what’s left to compete on? That’s the next one.
Not one part of that morning was a speed problem. Not one part of this is either.