Six months ago I had a clean theory about where AI was taking vertical software.

Pick a niche. Trade associations, say, or accounting firms. Own the customer relationship and the domain knowledge, then expand into every function they run. Start with one workflow, then take over their finance, their marketing, their member communications, their compliance. Knowledge of the customer plus distribution would be the moat, and AI would make it cheap enough to build everything.

That could still happen. I wouldn't bet against it. But I've started to see the opportunity differently, and I think the difference matters for anyone building, buying, or investing in software right now.

My thesis is that services and software are collapsing into each other. Software is learning to serve. Services are learning to scale.

Filing cabinets with buttons

Start with what SaaS actually was.

The first wave was horizontal. Salesforce gave every company a place to store customer records and a few workflows to move them around. HR systems, financial systems, and project tools followed the same pattern: a database, some forms, some automations, and a login.

The second wave was vertical. A CRM for dentists. Practice management for accounting firms. Software for contractors, for gyms, for associations. Better labels and a closer fit, but under the hood it was the same product. Storage plus a couple of workflows.

For twenty years, the software industry sold us filing cabinets with buttons on them. The software held the information. Your people still did the work.

And every customer bent to fit the product. You paid for two hundred features and used eleven. You hired an admin to babysit the system. You changed your process because the software couldn't change for you.

This was economics. Engineering time was expensive, so you built one product and sold it to ten thousand customers. Tailoring it for any single one of them never penciled out. Vertical SaaS was the best compromise available: generic enough to scale, specific enough to sell.

What changed

Two things.

First, shipping got cheap. The cost of producing working software has fallen so far that the constraint has moved. Building a feature for one customer used to be a bad trade. Now it can be a good one.

Second, the software can do the work itself. A model that can read, reason, draft, reconcile, and follow up means software can now perform the work it used to merely record.

Put those together and you get something SaaS never was: a capability. Software that takes on a function inside a company, gets shaped to how that company actually operates, and gets better at the job over time.

The delivery mechanism is the forward deployed engineer. The vendor sends people who sit with the client, learn the work, and tailor the capability until it fits. The customer gets far more than a login and a help center. That's a services posture. Software is learning to serve.

Two loops

Here's the part I find most exciting, and the part that separates a capability from a product. A capability compounds in two places.

Inside each client, it gets better as it learns that company. Its data, its exceptions, its preferences, the way this particular finance team closes the month. Day 300 is meaningfully better than day 1.

Across clients, the vendor gets better at deploying it. The tenth deployment goes faster than the first. The fiftieth surfaces patterns nobody could have seen at five. Every engagement teaches the next one.

SaaS had neither loop. The product on day 1,000 was the same as day 1, plus whatever features the vendor shipped to everyone at once.

The consulting trap

I ran a consulting firm for 18 years, so I know exactly where this can go wrong.

Forward deployed engineering is also how you build a services business with a software multiple on the pitch deck and a services margin on the P&L. Custom work feels like progress. Often it doesn't compound at all. Every client becomes a snowflake, every engineer becomes a bottleneck, and growth means hiring.

The test: does client eleven cost less, and get more, than client ten? If the answer is yes, and keeps being yes, you have a capability. If not, you have a consulting firm with a nicer website.

Services are learning to scale

Now look at it from the other side.

Professional services firms have always sold expertise by the hour. Expertise was valuable, and it didn't scale. Every engagement started close to zero.

The tempting AI move for these firms is to encode their expertise: write down the rules, build the playbooks into agents, bottle what the partners know. As I wrote earlier this month in Your Expertise Isn't the Moat, the Bitter Lesson puts a ceiling on that strategy. Hand-encoded expertise gets overtaken by models that keep getting more capable.

What does scale is an expert human working with an increasingly capable model. The model brings the scale. The expert brings the judgment. Every model release makes that pairing more powerful, and the data from every engagement makes it sharper.

That's services learning to scale.

Who wins

Follow both trends far enough and they meet in the same place: a company that sells a capability. Software does most of the work. Expert people steer it, tailor it, and stand behind the result. The whole thing gets better with every client.

Some software companies will get there by sending experts into the field. Some services firms will get there by putting increasingly capable models in their experts' hands. The winner is whoever owns the compounding loop: the deployment knowledge, the data from running the capability, and the judgment to keep improving it.

The losers are easy to spot too. Software companies still selling filing cabinets. Firms still selling hours.

What this means for you

If you buy software, stop grading feature lists. Ask what the vendor learns from deploying with you, and whether that makes your capability better next quarter. Underwrite the loop.

If you build software, your deployment engine matters more than your roadmap. How fast you tailor, how much each deployment teaches you, and how much of that learning carries to the next client.

If you run a services firm, your experts are about to get far more leverage. The question is whether you capture what they learn or let it walk out the door at the end of every engagement.

If you invest, look for the loop. A company where each customer makes the next one cheaper and better is a different animal from one that grows by adding headcount.

Software is learning to serve. Services are learning to scale. The companies that do both will look like nothing we've seen in either category.

Alan