What a buyer should notice when every tool adds AI

The news from productivity-software makers today runs on a single current: AI is now the thing that funding, onboarding and governance all orient around. For a business choosing tools, that shifts the useful question away from whether a product has AI and towards three plainer tests — does it stay up, does it stay under your control, and do the people building it understand the work.

None of those tests is new. They are the same questions you would ask of a filing cabinet or a bookkeeper. What has changed is that the marketing noise around AI makes them easier to forget, and the stakes of forgetting are higher when a tool can act on your data rather than merely store it. The items below are the ones from today worth a buyer's attention, in the order they matter.

Reliability is a feature, and it is measurable

The most useful thing a vendor can publish is an honest account of when it was down. GitHub's availability report for July states plainly that "In July, we experienced eight incidents that resulted in degraded performance across GitHub services" [1]. That is not a comfortable sentence to publish, and that is exactly why it is worth reading.

When you depend on a tool, its uptime becomes your uptime. An outage in a service you build on is felt by your customers as an outage in you, and you have no console to fix it with. So the question to ask of any vendor is not "do you ever go down" — everyone does — but "will you tell me when you do, and in what detail." A supplier that counts its incidents in public is giving you the information you need to plan around it: redundancy, a manual fallback, a status page you can point your own customers to. A supplier that says nothing is not more reliable. It is just quieter.

This is one of the quiet marks of a serious platform, and it is worth weighing before you commit. We wrote about the broader version of this in what enterprise-grade should mean to a small business: the grown-up features are rarely the flashy ones.

Deciding who — and what — may change your work

The second story is about control. GitHub's piece on AI contributors is addressed to the people who maintain open-source projects, but the lesson travels. Its summary notes that "AutoGPT maintainer Nicholas Tindle shares the repo instructions, gates, and boundaries that keep maintainers in control" [2].

Hold on to those three words: instructions, gates, boundaries. An AI agent that can submit work to your project — or, inside a business tool, draft an email, change a record, or move money — is a contributor you did not interview. The healthy response is not to ban it and it is not to wave it through. It is to decide in advance what it may touch, what it must ask about first, and where a human has to sign off. That is a configuration question, and a tool that lets you answer it is giving you something a tool that simply "has AI" is not.

This is the practical heart of the matter we covered in decisions automation should never make. The useful line is rarely drawn at the technology. It is drawn at the stakes of being wrong.

The people behind the product

SaaStr turned today to a less technical but equally practical buying signal: who is actually running the company you are about to depend on. The post opens by noting the author "wrote a version of this post a while back, and it struck a chord" [3], because founders kept recognising a pattern in the senior people they had hired into AI-era businesses.

For a buyer, the point is not gossip about executives. It is that the maturity of a vendor's leadership shows up, eventually, in the product: in whether support answers, whether the roadmap is coherent, whether promises made in a sales call survive contact with the invoice. You cannot audit a company's org chart before you sign. But you can read how it talks about its own work, how it handles its mistakes — see the first story — and whether its claims are specific or merely confident. Vague confidence from a supplier is a cost you pay later.

Where the money is going

Two items today show how much capital is chasing AI for business. Thrive Holdings, backed by OpenAI, "has raised $2 billion in new funding at a $12 billion valuation from investors like SoftBank, D1 Capital Partners, and Altimeter Capital" [4]. Numbers like that tell a buyer something real and something misleading at once.

The real part: AI tooling for ordinary businesses is where serious money now expects returns, which means more products, faster, aimed at smaller companies than before. The misleading part: a large valuation is a bet on the future, not a verdict on the product in front of you today. A well-funded vendor can afford to keep the lights on, which matters. A well-funded vendor can also afford to change direction, deprecate the feature you bought it for, or price up once you are committed. Funding buys runway. It does not buy fit. Judge the tool against your own work, which is the argument we made in choose software worth using.

Testing the code that AI now writes

The last item closes the loop. Blacksmith, which works on AI code-testing, reports that "revenue has grown more than tenfold over the past year" [5]. That growth is itself a signal about the state of the market: when AI writes a great deal of software, something has to check that the software is correct, and businesses are paying for that something.

The general principle outlives the specific company. Anything that generates work automatically needs a matching way to verify the work before it reaches a customer. A draft is not a decision; a generated record is not a confirmed one; code an agent wrote is not code anyone has read. The faster your tools produce output, the more it matters that you have a clear step where output becomes trusted. If a product offers you speed with no visible place to check its work, that is a trade-off to make on purpose, not by default.

The thread, pulled together

Five stories, one shape. Capital is flooding in [4], the volume of AI-generated work is rising fast enough to spawn an industry that checks it [5], onboarding is being written for first-time AI users, and the mature voices in the field are talking about reliability [1], control [2] and leadership [3]. For a business choosing tools this week, the quiet news is the reassuring part: the questions that have always separated a dependable supplier from a loud one still apply. Does it stay up. Can you draw the line on what it may do. And do the people building it sound like they have done the work before.