A feature gets approved with a budget behind it. Months later it turns up on your desk as: here's the requirement, you've got two days.
I used to assume that was just how it goes, the price of working somewhere big enough to need a budget at all. But I kept noticing how much time went missing between the approval and the requirement, and how little of that time involved anyone actually building anything. Where does it go, exactly.
Maybe it goes to something as unglamorous as coordination cost. Not a phrase anyone puts on a budget line, but if you added up every hour spent negotiating, clarifying, chasing a reply, translating one team's version of a plan into another's, I suspect it would be the biggest line item nobody tracks.
Information isn't the bottleneck
Here's a distinction I keep coming back to, half convinced it's obvious and half convinced nobody actually acts on it.
The internet solved an information problem. Search, dashboards, SaaS, wikis, all of it made data cheaper to find and share. What it never quite managed was making organisations any faster at agreeing on what to do with that data.
Confluence, Jira, Slack, Power BI: each one gave us better access to information. None of them answered the question that actually slows a day down, which is closer to "what do I need to know right now, and who do I have to ask." A person still had to close that gap. Chasing approvals. Negotiating priorities. Explaining, again, why the thing that made sense in one meeting doesn't quite make sense in another.
That's a fair chunk of what a manager's day looks like, I think. Not because organisations are fond of hierarchy for its own sake, but because someone has to absorb the cost of all that translating, and it might as well be someone whose job description says so.
What AI actually changes
This is where I think AI is doing something the internet never did, though I'll admit the evidence is more "things I've noticed" than "things I've proven."
It's not just handing you more information and leaving you to sort it out. It's doing some of the sorting. Ask the old way whether a piece of software is approved for use, and you're filing a ticket, pinging someone in IT, waiting for a reply that may or may not exist yet. Ask an AI system that can see the policy docs, the approval records, the relevant Slack threads, and the answer just turns up. Nobody's interrupted. Nobody has to remember to reply.
Preparing for stand-up used to mean scrolling through yesterday's thread, the pull requests, the deploy log, and stitching a coherent story out of it yourself, usually at 8:52am. Increasingly, that story is just sitting there when you open the tab.
None of that sounds like much on its own. Small, slightly boring wins. But stack enough of those small wins on top of each other across a week, and something starts to shift. The internet digitised our data. AI might be quietly starting on our conversations.
Everyone gets a little bit of manager
Here's the idea I can't quite let go of, even though I'm still turning it over.
A lot of what a manager does, day to day, isn't really judgement. It's routing: gathering an update here, chasing a status there, translating a decision out of one team's dialect and into another's. AI turns out to be reasonably good at exactly that kind of work, which raises an odd question. If the routing doesn't need a person anymore, does it need to sit with the manager specifically?
I don't think managers disappear. But some of what used to be theirs alone starts leaking outward to whoever's underneath them. A junior engineer can get a design reviewed, a risk flagged, a stakeholder update drafted, without waiting for someone more senior to find twenty free minutes. Not because the AI has better judgement than they do. Just because it's harder to claim you didn't have a way to find out.
"I wasn't told" was always a slightly convenient answer. It's about to get a lot less convenient.
Management edges toward leadership. Once the update-gathering stops being anyone's whole job, what's left is the part that was always the hard part anyway: coaching someone through a bad week, making a call nobody wants to make, having the conversation that keeps getting put off.
What probably doesn't change
I want to be careful here, because it would be easy to turn this into a bigger claim than I actually believe.
Somebody still has to own the decision. AI can be wrong with total, cheerful confidence, and when it is, the accountability doesn't evaporate along with the mistake. Banks, hospitals, anything with real regulation attached: the paperwork exists because the risk is real, not because nobody thought to skip it yet. Cheaper coordination doesn't make being wrong any cheaper.
And I'd guess not everyone wants the extra autonomy this implies. Plenty of people are genuinely happier with a job that has clear edges, and I don't think that's a failure to keep up with anything. It's just a preference, and a reasonable one.
What I do think shifts is the default. When the answer is one question away, "nobody told me" stops being a place a sentence can comfortably end. The frontline edges a little closer to the decision. Not because anyone redrew the org chart on purpose. Just because the distance between having a thought and doing something about it got a bit shorter than it used to be.