Agencies sell deliverables. A lot of agencies are now focused on making those deliverables faster and cheaper with AI.
“We reduced blog posts from three hours to twenty minutes.”
Speed and price are not nothing. But you’re hurling to the speed of instant and price of zero and optimizing for them this way is a dangerous distraction.
The first agency to make something 10x faster gets a margin gift. The tenth agency gets a new baseline. The hundredth agency gets a client asking why it still costs so much.
That is the value decay of compressing delivery. You do not just suddenly have the same business with better margins. You get a different mix of value, new bottlenecks, and a decay curve that is bad and you can’t know how bad.
Reducing cost while maintaining or slightly improving quality is rational. Especially when no one knows what the fuck to do with all this uncertainty and change.
I often feel like a rock sitting in the middle of a river and work and life is the water just gushing by. Or maybe I’m a snail on the rock waiting to get picked off by a bird.
If there is downward pressure in the market, you kind of have to participate right? Ignoring delivery compression would be its own kind of denial.
But it cannot become your main focus.
At best, it buys time. And not in a clean “we have 18 months to figure this out” way. More like the pot of water gets a little warmer every quarter but margins still look okay… except for that Claude token burn.
Let’s say you are a content-heavy agency and you find 50% efficiency gains in production. Great. What now? Are you going to
- Jack Dorsey your company? cut 50% of your writers?
- reduce client pricing by 50% before the market forces you to?
- increase output 5x and hope clients also think more of the same is better?
- keep the margin then wait and see?
To me every answer here sucks and has a long-term cost that far exceeds the short-term gains. Whether its financial, org, trust, team, brand. Who is still in going to be in the foxhole with you after whatever you decide to do with that gain?
There is another option I think. Maybe, naive. If the productivity gains per human can be on average 100x’d over the next two to three years, then cutting just 10 people now is the worst decision you could make. You’re not going to “learn how to” reskill new people later. Where people fit is the number one question you should be asking after you compress delivery.
The trap is that compression feels like winning
Delivery compression is a bad center of gravity. It can give you a little more margin and a little more time. But if it becomes the focus, and I see people bragging about slop gains all day… it’s kind of addictive right? It feels like progress and so natural next question is, “How do we compress again?”
Yes, it is unavoidable, but it is not your North Star if you want to be around in two years.
And the trap isn’t just that the curve decays. It’s that compression rewards you for getting better at the thing that’s becoming worthless. Every “we got it down to 20 minutes” newsletter or LI announcement is a hit. The scoreboard is legible.
Decision-layer work has none of that, no before/after screenshot, no number to brag about, no post that performs. By the time the margin gift is gone, you’ve spent two years training your team, your clients, and your own attention on the wrong scoreboard.
And you lose the muscle to do the harder work, because everyone was automating the old ways of delivery.
Where can you grow with a compression strategy?
You can add new services. But you have no human SMEs to tune pipelines or evaluate agent outputs. I have this problem right now. I have a team of 10… all are engineers by trade.
You can lean on AI to create a market research agent, which your client can also do for $20/month about 90% to 110% as well depending on their context.
That’s why people say really dumb things like “the new frontier model is so good at writing content.” Gets coffee, cracks open Claude, sets it to Opus 4.7. Prompt: “flesh out the attached concept into an article” Response: <slop>.
Adding services can just create more delivery surface area. Cutting price might protect retention for a while, but it trains the market to value you less. Increasing impact is the right answer, but it is also the hardest one because it forces you to confront what you can actually prove.
You can increase impact. You can reduce the cost for clients to work with you before they force the issue. Maybe you can do all three. But only one of those is a real answer.
And the current thought-leader answer to impact getting harder to measure is not good enough.
The zero-click marketing version is basically: stop caring so much about traffic, get comfortable with ambiguity, and infer attribution through softer signals.
I get why people are saying it. Some of it is directionally true. But it is not a satisfying answer for an agency that has to get renewed next year.
I heard a version of this recently, and it felt like the same talk from two years ago with new AI anxiety wrapped around it. The recommended attribution approach only really works if you run one clean campaign or experiment at a time against a leading indicator. That is not how most client environments work. It is not how most agencies operate. It is not how messy markets behave.
So no, the answer is not to convince clients to care less about measurement.
And it is not to sell them more blog posts because blog posts got cheaper to create.
That combination is the worst of both worlds: more output, more risk, weaker attribution, and less defensibility when someone higher up asks what any of it did.
Compression crowds out the harder work
The bias toward compression crowds out the harder work.
Instead of asking, “What needs to change in our model now that delivery is cheaper?” agencies ask, “Where else can we save time?”
That is understandable. I have the same bias. I like building. If I can make an app in one-hundredth of the time, my first instinct is to make 100 apps. That instinct is great when I’m building Magnetiles with my kids. In business, it is usually dangerous.
Because the fact that you can make more does not mean more is the answer.
More production creates more questions around sales, trust, measurement, client education, positioning, renewal risk, and proof of impact. Those are not side issues. They are the business.
If a client can buy more output for less money, the agency has to answer a much harder question: why should they keep buying from you at all?
Not because you can produce.
Not because you can produce faster.
Not because you can produce more.
Because you help them make better decisions about what should exist, what should not, what is worth measuring, what is too noisy to trust, and what the next move should be.
How decisions get made is the product
That is the layer agencies have to move into.
How decisions get made is the product.
That does not mean execution stops mattering. It means execution becomes downstream of a better decision system. The agency still has to ship good work. But the work is no longer the main source of defensibility.
The defensibility is in the decision layer around it.
- What should we do next?
- What should we not do even though it is now cheap?
- What signal deserves trust?
- What metric will create the wrong client conversation?
- What work increases renewal odds?
- What work just increases output?
- What has to be true before this recommendation is safe?
- What did we learn last time that should change the system this time?
That is where scar tissue matters.
Experienced agencies have a lot of scars. They have learned what happens when a report is technically correct but politically useless. They have learned what happens when a recommendation has no owner. They have learned what happens when a client buys activity but renews on confidence.
But scars only matter if they become negative knowledge.
A scar says, “That hurt.”
Negative knowledge says, “Do not let this happen again under these conditions.”
That is the move from experience to infrastructure.
And that is the work I think agencies are underestimating. The point is not to automate the old model until the margins disappear. The point is to convert hard-won judgment into a system that makes better decisions than the old model could.
Delivery compression is real. You should use it.
But do not confuse it for transformation.
It is a temporary advantage on a decaying curve.
The durable work is deciding what deserves to exist in the first place.
The case that makes this confusing
[Working section — decide whether this lives here, becomes its own post, or gets cut]
The hardest version of this argument to make is when you can point to an agency that’s doing the opposite of what I’m saying and clearly winning.
There’s a top agency in my area, seemingly very innovative over the past ten years, focused mainly on the infrastructure of formalizing compression. They’ll probably be fine. Reputation compounds, the marketing engine works, procurement defaults keep holding, and clients are afraid to not just pick them. But under all of it, the offer is weak: an inch of compression in every direction.
They’re not wrong in any quarter. Every internal review shows the compression program delivering. Nobody at that agency is going to stand up and say “our offer is weak” because by every metric they track, it isn’t.
But what they’ve industrialized is the thing clients will eventually do themselves or buy from someone cheaper. The moat is reputation and switching cost, not the work. That’s a real moat — for a while. Reputation moats erode quietly. They don’t crack, they just stop converting. RFP shortlists get longer. Renewal conversations get more pointed. New logos skew smaller or weirder. From the inside it looks like a soft quarter, then another one.
What they’re actually optimizing is defensibility of the current book, not defensibility of the category position. Those look identical until they don’t. And the tell is exactly that — an inch of compression in every direction. That’s a portfolio strategy for a business that has decided the game is efficiency. It’s coherent. It’s just a bet that the decision layer never becomes the thing clients buy.
The uncomfortable part isn’t whether they’re right or wrong. It’s that they can be wrong for five years and still print money, which makes them a confusing benchmark. The risk isn’t that they beat you. It’s that they make the wrong game look like the only game.
