AI can connect the silos. You still have to learn how people work

AI is making it much easier to see how work connects across teams. It has not made the people part of the work any less important.

I’ve spent a lot of this week moving between pieces of work that could easily have stayed separate in a conventional content organisation: campaign planning, editorial briefs, podcast questions, GEO research, event reporting, customer evidence, video, web production and stakeholder approvals.

Different people own different pieces. Some are content specialists and some are not. Some are structured, some are messy, some give very clear updates and some only surface when they need something. The work itself is connected, but the organisation rarely presents it that way.

AI has made those connections much easier for me to see. It has also made something else clearer: the better AI gets, the more valuable it becomes to understand how particular people actually work.

AI is good at making the whole system visible

One of the biggest changes in my own work is that I can now hold much more surrounding context at once. I can take a campaign plan, a content audit, ten briefs, meeting notes, audience pain points, publishing timings and a separate evidence-gathering plan and ask how they fit together.

That used to be difficult partly because the information lived in different places. A deck over here, a transcript over there, a spreadsheet from another team, a note from a meeting three weeks ago, a Workfront comment, a message in Teams. The connections existed, but reconstructing them manually took time.

AI reduces that cost dramatically. This week, for example, I could see that a campaign’s editorial recommendations, paid-media flighting, audience pain points, roundtable evidence and Sage Advice production schedule were all parts of the same decision.

That sounds obvious once written down. It did not arrive that way. It arrived as separate pieces of work owned by different people.

Seeing everything is not the same as knowing what to do with it

That campaign work also exposed the limitation.

The initial analysis was comprehensive. Too comprehensive. There were 31 possible editorial recommendations, ten had been prioritised, several overlapped, some fitted the early campaign phases and others made more sense months later. Some depended on activation timing and some could benefit from evidence that had not been gathered yet.

It would have been easy to turn that into an enormous production plan. AI is very good at doing that. Give it enough context and it can produce a table with every recommendation, dependency, date, stakeholder and rationale.

The problem is that the person who ultimately needed to review the plan was not an editorial strategist. She did not need the whole decision tree. She needed something closer to four articles I recommend now, one that depends on the November activation, one supporting piece we can add if it is useful, a later pipeline we deliberately do not lock yet, and five campaign decisions I need before production can be finalised.

The useful work was not adding more analysis. It was removing it.

That is partly experience

I have probably become more sensitive to this over time. You learn what happens when you send somebody 25 options and call it a recommendation. You learn which stakeholders want to see the reasoning and which primarily want to know what you think they should do.

You also learn that different people need different interfaces to the same work. Someone may be very strong on campaign context but not want to make granular editorial decisions. A video producer may need the questions that affect capture, not the whole content strategy. A web producer needs copy that fits the component, not another strategy deck.

Some people are poor at updates. Some are excellent at producing raw material but need help narrowing it. Some respond well to a rough idea and others need something almost finished before they can react properly.

None of this appears on the org chart, but it determines whether work moves.

It is also empathy

There is a small piece of perspective-taking involved every time you hand work to somebody else. What does this person already know? What decision are they actually qualified to make? What would overwhelm them? What would make them feel I had simply handed my problem to them?

I saw the same thing while working on podcast questions. It would have been easy to take the GEO opportunity and turn a normal interview into a giant research instrument: more questions, more follow-ups, more evidence gaps, more possible outputs.

Again, AI could help create all of that. But the person conducting the interview still needs a conversation they can actually use. So the job becomes compression: keep the questions that might unlock something valuable, remove the ones that make the interview feel engineered around the internal content system.

The system can be complicated behind the scenes. The experience for the next person should not have to be.

AI abundance makes restraint more valuable

For years, a lot of knowledge work was constrained by production. Could we research enough? Could we generate enough ideas? Could we write enough options? Could we analyse all the material?

AI weakens many of those constraints. The problem increasingly becomes the opposite. We can generate too much: too many recommendations, too many questions, too many possible content angles, too many dependencies and too many perfectly plausible plans.

That makes restraint more valuable. The scarce skill becomes knowing what this particular person needs, in this situation, at this moment, to move the work forward.

That is not prompt engineering. It is judgement built through working with people.

AI is still changing the shape of the organisation around the work

There is another side to this. AI has made it much easier for me to work across groups that would normally remain fairly siloed.

A piece of evidence gathered at an event can now be connected quickly to a GEO question, a campaign theme, a future article, a video opportunity and a customer-evidence gap. A conversation in one project can expose a pattern that matters somewhere else. A campaign brief can be compared against earlier research without somebody manually holding both documents in their head.

In that sense, AI is becoming a connective layer across work. That is powerful, and it is one reason I keep coming back to prompting as routing: the interesting problem is increasingly not generating a thing, but deciding where context should go and what should happen next.

But connections still have to be routed through people. The campaign team needs one version, the content team needs another, the video producer needs the questions that affect capture, and the web team needs the copy that fits the component. The stakeholder who is not a content specialist needs the recommendation, not the entire editorial reasoning chain.

AI helps expose the system. People still experience the system through other people.

Maybe this is part of what seniority actually is

I used to think moving into more senior work meant doing less of the detailed work and more strategy. I am less sure that is the useful distinction.

A lot of the work I now think of as senior looks more like seeing the connections, understanding the constraints, knowing where a decision actually belongs, reducing complexity before it reaches somebody else, and giving different people different views of the same underlying work.

It also means keeping specialists involved without making everybody responsible for everything. Sometimes the senior move is not another meeting, another framework or another 20-row plan. Sometimes it is turning 31 possibilities into four recommendations.

This connects with what I wrote in The content job is moving up a level. The work can stay close to writing, interviewing and evidence while responsibility expands around the system that makes those things useful.

The brightest AI will not make this disappear

AI will get better at modelling stakeholders. It will remember preferences, learn communication patterns and probably become very good at predicting how a particular person is likely to respond.

I still do not think that removes the need to learn how people work, because working relationships are not static datasets. Someone may want more detail because a project is politically sensitive. A normally decisive stakeholder may become cautious because they are new to the subject. A team that usually moves quickly may suddenly have no capacity. A person who rarely needs context may need it because the decision itself has changed.

You learn these things by being in the work with people, then you adapt.

AI can help me make the connections. Experience helps me understand the people those connections have to pass through. Increasingly, I think that is where a lot of the real editorial intelligence sits.

The system may be AI-assisted. The work still has to be made usable for humans.