What the software you buy is building, checking, and keeping

Today's reports from productivity-software makers share a quiet theme: the interesting question about AI has moved from whether to adopt it to what happens once you do. A survey of AI builders, a code-review experiment, a vertical-software milestone, a churn primer, and a small interoperability change all sit on the same practical ground — how tools connect, how their output is checked, and whether the customers who arrive decide to stay.

For a business choosing tools, that ground matters more than the feature lists. A feature is a promise about a good day. The things below are about ordinary days: the maturity of the vendor you are buying from, the quality of the work an agent produces when nobody is watching, the retention that decides whether any of it compounds. None of these is glamorous. All of them decide whether a purchase was wise a year later.

The builders' economy, in numbers

A "State of AI" report is easy to skim past as an industry ritual. It is more useful read as a mirror held up to the companies that make your software. When those vendors are themselves rushing to add AI, their maturity — how carefully, how expensively, how repeatably they build — becomes part of your risk when you sign up.

ICONIQ published the third edition of its report today, and the framing is worth noting. "ICONIQ just released the third edition of its State of AI report, subtitled 'The Builder's Economy,' built on a Q2 2026 survey of roughly 305 executives at software companies building AI products." [1] The phrase "builders' economy" is doing real work. It says the centre of gravity has shifted from companies that talk about AI to companies whose product *is* an AI system they have to run every day.

What should a buyer take from that. Chiefly a question to ask any vendor: is the AI in this product a durable capability the company can maintain, or a feature bolted on to keep pace. The two look identical in a demo and diverge sharply in year two. If you want a longer treatment of where automation earns its place and where it does not, we have written about what AI should and should not do in your business.

When more capable tools change the result

There is a comfortable assumption in software that better tools produce better output. GitHub reported today on a case where that assumption did not hold cleanly, and the lesson generalises well beyond code.

The report describes "how migrating Copilot code review to shared Unix-style code exploration tools reduced review cost by reshaping agent workflows around pull request evidence." [2] Read that slowly. The improvement did not come from a more powerful model or a longer feature list. It came from anchoring the work to concrete evidence — the actual contents of a change — and simplifying the tools around that.

The trade-off worth carrying into your own decisions is this. When you hand a task to an automated agent, giving it more capability is not free. More options can widen what the agent considers and blur what it focuses on. The teams that get useful results tend to narrow the job, tie it to specific evidence, and check the output against something concrete rather than trusting the tool's confidence. That is a design discipline, not a product feature, and it is one you should look for — or impose — whenever you let software act on your behalf.

Vertical software crosses a milestone

Most tools you evaluate are horizontal: a spreadsheet, a general CRM, a chat app that any business could use. Vertical software is the opposite — built for one industry, shaped around how that industry actually works. ServiceTitan is the standing example, and today it crossed a threshold worth understanding.

SaaStr's account is plain about what the company sells. "It sells an end-to-end operating system to the trades: HVAC, plumbing, electrical, roofing, garage doors." [3] The headline numbers — past $1B in ARR, still growing 25%, net revenue retention at 110% — describe a business whose customers not only stay but spend more over time.

The lesson for a buyer is about fit, not size. A vertical tool that knows your trade will ask for less configuration and fight you less often, because its defaults already assume how your work runs. A horizontal tool is more flexible and less opinionated, which cuts both ways. Neither is simply better. The right question is how much of the tool's built-in assumptions match your reality, because every gap becomes work you do by hand. That is the same test we apply in choose software worth using: does it reduce the work, or move it.

Churn is the number under everything

The most useful piece today is also the least new. SaaStr ran a reader question on reducing churn, and its opening is a sentence every operator should keep close: "Churn is one of the most critical metrics in software." [4]

Churn is the rate at which customers leave. Its power is arithmetic. Growth compounds only on the base that stays, so a business that loses customers steadily has to run harder every month just to stand still. This is why churn is described as a silent killer — it rarely shows up as a crisis, only as growth that never quite accelerates.

For a business choosing tools, churn cuts two ways. It is a metric to watch in your own customers, and it is a lens on the vendors you buy from: a tool with high churn is a tool people found reasons to leave, and their reasons may become yours. Reducing it is mostly unglamorous work — noticing which customers have gone quiet, replying before they drift, keeping the record current so nobody has to ask twice. We have argued before that follow-up is the whole job, and churn is the number that proves it.

A small change to how systems connect

The last item is the smallest and, in its way, the most practical. Google Workspace's weekly recap notes a new interoperability route: "You can now join video conferences on Google Meet hardware via SIP through a Pexip interop gateway." [5]

SIP is a common standard for connecting calling systems. The point is not the acronym. The point is that a room built around one meeting platform can now join meetings hosted on another without new hardware. That is what interoperability buys you — the freedom to not have everything come from one supplier, and the ability to work with partners who chose differently.

This is the least dramatic entry today and the one that touches the most ordinary frustration: tools that will not talk to each other. Every standard supported, every gateway opened, is one less place where your work stops because two systems could not agree. We have written at length about why your tools do not talk to each other, and why a small interoperability change is often worth more than a headline feature.

What ties the day together

Five reports, one thread. The builders' economy tells you the companies you buy from are themselves mid-build. The code-review experiment shows that capability without discipline does not produce quality. The vertical-software milestone is a lesson in fit. The churn primer is the reminder that keeping is harder and more valuable than getting. The interoperability note is proof that connection still beats features. A business choosing tools this week could do worse than ask, of every option on the table, five plain questions drawn from those reports: is the vendor building this to last, is the output checked against something real, does it fit how we actually work, will people stay, and will it talk to what we already own.