There’s a trend running through agencies right now that I find genuinely interesting and slightly worrying, in equal measure.
Vibe coding. Automation of everything.
AI tools have made it possible for people (like me!) who’ve never written a line of code to build working tools, connect systems together and automate chunks of their business in an afternoon.
On one hand, this is brilliant. Coding has been opened up to non-coders. Things that would have needed a developer and a budget two years ago can now be knocked together by anyone with a spare hour and a rough idea. Some of what I see founders building for their own operations is genuinely useful.
On the other hand, when building things becomes that easy, we start building things that don’t need to exist. Automating processes nobody was struggling with. Adding tools that create more complexity than they remove. The ease is the trap: when the answer to “could we automate this?” is always yes, nobody remembers to ask whether we should.
And somewhere in that rush, I keep seeing agencies drift towards automating the one thing they really shouldn’t.
The thing the client is actually paying for.
The pressure to automate everything (and why it’s not silly to feel it)
First, some fairness, because the pressure on agency founders and teams here is real.
Your clients are asking what you’re doing with AI. Your competitors are shouting about their AI-powered processes. Every conference talk and LinkedIn feed says the agencies that don’t go AI-first will be left behind. And your margins are under enough pressure that anything promising efficiency deserves a look.
So no, it isn’t silly that “automate more” has become the default direction of travel. If everyone around you is treating efficiency as the scoreboard, you’d be odd not to.
The problem isn’t automating systems or processes. I’m not here to tell you AI is overhyped or that the tools don’t work, because plenty of them do. The problem is automating without ever answering one question first.
What is your client actually buying?
Here’s the question and it’s worth actually taking a moment to think about rather than rushing past.
When a client pays your agency, what are they actually paying for?
The obvious answer is the deliverables and results. The strategy deck, the report, the campaign, the copy, the build. But I don’t think that’s right and I don’t think it’s ever been right.
The deliverable is the receipt.
What they’re actually buying is the judgement that produced it. Knowing which thing to do and why. Knowing what not to do. The taste to tell the difference between work that’s technically fine and work that’s good. The experience to spot the problem behind the problem they brought you.
That distinction didn’t matter much before, because the deliverable and the judgement always arrived together. AI has split them apart. It can now produce the deliverable, at speed, to a standard that often looks fine.
But are you okay with being fine?
If AI produces the work and nobody senior has exercised real judgement over it, your client is paying agency prices for something they could prompt their way to themselves. And increasingly, they know it.
They ask – what value are you adding?
Automate the prep, protect the product
So here’s the split I’d actually advise and it’s not complicated.
Automate the mechanical layer freely. The admin, the reporting plumbing, the research gathering, the meeting notes, the first pass at anything formulaic, the internal ops. None of that is what anyone’s paying you for. Every hour AI saves you there is a clean win and honestly, agencies that aren’t doing this are leaving margin on the table.
Then. protect the judgement layer. What to do, why, in what order, to what standard and what to leave out. The senior read on whether the work is actually good. The conversation where you tell a client the thing they asked for isn’t the thing they need. That layer is the product. Handing it over to AI isn’t being efficient, it’s giving away the reason you can charge what you charge.
The skill in this era isn’t using the most AI. It’s knowing exactly what not to hand over.
The headspace trap (where the saved time actually goes)
There’s a second layer to this that I think gets missed entirely and it’s a view I’ve been sitting with for a while: AI saves time, but it doesn’t save headspace.
(Anyone who’s worked with me will know how much I talk about headspace as a factor in being productive and effective!)
The promise is that automation frees you up for higher-value thinking. In practice, I watch the freed hours get refilled almost instantly with more throughput. More clients, more output, more volume, at the same pace, with the same tired founder running it. The calendar changed shape. The mental load didn’t.
Which connects the two halves of this piece. The recovered time has one genuinely valuable use: the judgement work. The thinking, the taste, the client counsel. That’s the layer AI can’t do, the layer clients are paying for and the layer that most needs an unhurried human brain.
So the founders getting this right aren’t the ones who’ve automated the most. They’re the ones using what automation gives back to get better at the bit that can’t be automated.
What to actually do (a one-hour exercise)
If you want to test where you stand, this takes about an hour.
1. Write down what clients say they pay you for, in their words
Not your service list. Pull it from testimonials, renewal conversations, the reasons given when you win a pitch. It’ll be things like “they get our market”, “they tell us the truth”, “they make the complicated simple”. Notice how rarely it’s a deliverable.
If you have AI generated meeting notes stored somewhere of sales conversations, go and find them – the answers will be there and AI can (ironically) help you find them quickly.
2. List where AI currently sits in your delivery
Now, get a feel for where things are at right now in your agency. Go and find every step where a tool is drafting, producing or deciding something.
You’ll probably need to chat with your team to get the full picture here. Don’t overthink asking. Just send a quick email or Slack message to them and ask where they are using AI tools right now.
3. Put the lists side by side
Anywhere AI is producing something on the “why they pay us” list, ask: is a senior human genuinely exercising judgement over this, or skimming and shipping? Be honest about which one it is, because your clients will eventually be able to tell.
4. Pull back anything that fails
Not by removing the tool necessarily, but by putting the judgement back on top of it. Add the human judgement and taste back into the deliverable so that it becomes what the client is actually paying for.
When heavy automation is the right call
Okay there are exceptions to this.
Here are two of them.
- If your agency’s positioning is deliberately AI-enabled delivery at a lower price point, automating the product is the strategy. That’s a legitimate business, just a different one and it needs pricing and margins built for it.
- And production-heavy commodity work that clients treat as commodity anyway was always going to go this way; judgement was never what they were buying there.
The point isn’t that automation is dangerous. It’s that it should follow from what you’re actually selling, not from what the tools make easy.
Here’s the uncomfortable end of this. If you automate your core product, the efficiency gain is real, but so is the destination: work that’s indistinguishable from what your client could generate themselves. You’ll get there faster and cheaper than ever. It’s just not somewhere worth arriving.
The agencies that come through this era won’t be the ones that automated the most. They’ll be the ones that remained worth paying for because the value that they delivered couldn’t be replaced by some smart prompts or vibe coded tools.