I swore off agencies. AI is making me reconsider.

Everyone knows the old rule.

That trilemma used to govern almost every service business. Websites, content, operations, sales support, internal tools, reporting, customer service, lead generation, admin. If you wanted quality at speed, you hired expensive humans. If you wanted cheap work quickly, you accepted slop. If you wanted good work without paying through the nose, you waited.

That is the part I think people still underrate.

Most small and medium businesses do not need some grand AI strategy. They do not need a “chief AI officer.” They need someone to look at the messy parts of the business and ask: why is a human still doing this by hand?

Why is someone manually replying to the same leads? Why are sales calls disappearing into memory? Why are proposals written from scratch? Why is reporting a monthly archaeology project? Why are staff searching through WhatsApp, email, Google Drive, and one cursed spreadsheet to answer a basic customer question?

AI is useful because it attacks that layer. The dull middle of business. The part nobody puts in the pitch deck but everyone feels every day.

Why I Hated Agencies

I used to be in the agency business. Then I got tired of it.

Agency work can become a very elaborate way to sell custom promises, coordinate humans, chase approvals, absorb client chaos, and then repeat the whole thing again next month. A lot of it is low leverage. A lot of it is manual glue. A lot of it depends on heroic effort from people who eventually burn out.

And the worst version is when the client actually needs systems, but the agency is only structured to sell deliverables.

They need faster follow-up. They buy “marketing.” They need better sales ops. They buy “automation.” They need clearer data. They buy “dashboards.” Everyone is half right, which is why everyone is half frustrated.

So for a long time, I had zero interest in going back. I did not want another service business where the product was my calendar and the margin was my sanity.

What Changed

AI changed the leverage.

A sharp operator can now do what used to require a designer, developer, analyst, copywriter, ops person, and project manager sitting in a very expensive group chat. Not perfectly. Not magically. But enough to matter.

The work becomes more like implementation, systems design, and productized service delivery. Less “please make me content.” More “build me a machine that captures demand, follows up, drafts the reply, updates the CRM, flags edge cases, and teaches the team what is happening.”

That is exciting.

Not because AI agencies are automatically good. Most will be rubbish. Some will just be old agencies with ChatGPT in the brochure. But the best version of this category is genuinely interesting: small teams that go deep into real business workflows and ship working systems.

Even YC is shilling AI services now, which would have sounded illegal five years ago. But I think they are directionally right. When AI reduces labor cost and increases delivery speed, services can become more productized, more measurable, and more outcome-driven.

That is the fun version of the agency business. Not decks. Not retainers for vibes. Actual business improvement.

The Catch

The annoying thing is that AI is also moving too fast to package neatly.

As I am writing this, Claude Fable 5 has basically just landed. I have not even properly played with it yet. By the time you read this, it may already be old old news, replaced by another model, another agent framework, another coding workflow, another “wait, this changes everything” moment.

That makes AI very hard to productize.

In normal software, you can define the product, write the docs, build the sales motion, and improve slowly. In AI, the frontier keeps moving while you are still naming your packages. The best workflow last month can look primitive next month. The thing that was impossible on Monday becomes a demo on Friday and a commodity by the following quarter.

So the real value is not just knowing the tools. The value is getting hands-on with a business, understanding the ugly workflow, and translating new capability into something that actually moves a number.

And oh lord, do I not feel naturally inclined to do that.

The Love-Hate Part

That is my relationship with AI right now. Love-hate.

I love how useful the tools are. I love that one person can suddenly build, research, write, analyze, automate, and ship at a level that used to require a whole bench. I love that boring businesses can get capabilities that were previously locked behind enterprise budgets.

I hate how much the value comes from being in the mud. Sitting with the owner. Watching the real process. Seeing the weird exception. Understanding why the staff ignores the existing system. Finding the one manual step everyone knows is stupid but nobody has had time to fix.

But that is also exactly why this is valuable.

The opportunity is not “AI tools.” The opportunity is applied AI taste. Knowing what to automate, what not to automate, what to keep human, where errors matter, where speed matters, and where a small workflow change can create an outsized result.

What Businesses Should Do

My minimalist advice for business owners:

• Start with one painful workflow, not an AI roadmap.
• Pick work that is repetitive, text-heavy, rules-heavy, or follow-up-heavy.
• Keep a human approving outputs until the system earns trust.
• Measure one boring number: hours saved, leads replied to, response time, revenue recovered, mistakes reduced.
• Do not outsource understanding. Get your hands dirty enough to know what good looks like.

If you are a young business owner, this is even more important. Do not just consume AI content. Use the tools. Build the ugly internal thing. Automate one annoying admin job. Make the first version yourself, even if it is embarrassing.

The owners who win will be the ones who can look at a process and instantly feel where AI should go.

Why This Might Pull Me Back In

I have never worked at an outside company before. I have always built my own things, chased my own problems, and paid for my own mistakes.

But AI has created this weird new itch.

The idea of joining an already great company and helping take it to the next level suddenly sounds exciting. Not in a corporate ladder way. In a “give me a real machine, real constraints, real customers, real operations, and let me see where AI can bend the curve” way.

That is also why AI agencies are interesting again. A great AI agency is not just a vendor. It is a temporary operating system upgrade for the business. It brings taste, tools, implementation speed, and outside pattern recognition into a company that has real problems worth solving.

Fast. Gud. Cheap.

For the first time, maybe you actually get all three. Not everywhere. Not automatically. But often enough that the old rule is breaking.