You Can Build the Software You Need by Describing It
For as long as you’ve been in business, the deal on software has been the same. You rent something that almost fits, you work around the parts that don’t, and if you want the piece that actually fits your business you hire a developer and find out what that costs.
The second half of that arrangement came apart last month, and most owners haven’t heard about it because the coverage was written for technology companies.
You can now describe the thing you need in plain English and have it built.
Why It Matters
On July 7, Anthropic put Claude Cowork on the web and on phones. Two days later OpenAI launched ChatGPT Work. Same idea from both: instead of asking a question and getting an answer, you hand over a goal. It reads your files, pulls from whatever apps you’ve connected, works through the steps on its own for as long as the job takes, and gives you back finished material. Spreadsheets with working formulas. Slide decks. Documents. In some cases a small working web app.
That got covered as a race between two companies, which is the least interesting thing about it. The interesting part is who turned out to be using it.
Anthropic published an analysis of 1.2 million anonymized sessions from more than 600,000 organizations, taken over three weeks in May. Every session was sorted into one of twenty categories of work.
| What the work actually was | Share of sessions |
|---|---|
| Business process and operations | 33.4% |
| Content creation and copywriting | 16.4% |
| Software development | 8.7% |
| DevOps and infrastructure | 7.0% |
| Research | 6.4% |
| Data analysis | 5.8% |
More than 90% of it has nothing to do with writing software. The largest category by a wide margin is ordinary operational work — Anthropic’s own examples are pulling scattered updates into a single report, building an onboarding checklist, reconciling spreadsheets.
They have a phrase for it that’s worth borrowing. The work around the work. The jobs that are nobody’s actual role but attach themselves to every role, and that in a small business attach themselves to you.
These were built for programmers. The people who took them over are the ones doing admin.
Tomorrow’s forecast, in your inbox.
One email. Five minutes. Written for owners, not engineers.
What This Looks Like in a Real Business
I should say plainly that I’ve been working this way for a while, so this isn’t a report from a distance.
Three of the web tools running in my own business were built by describing what I wanted rather than by hiring anyone. The site you’re reading this on runs on a theme I had built the same way, revised more times in an afternoon than I’d have been able to request from a developer in a month. None of that required me to write code, and none of it required a budget.
What it did require was being able to say clearly what the thing should do. That turns out to be the whole skill, and it’s a skill you already have from years of briefing employees and contractors.
The practical shape of it for an owner: a spreadsheet that reconciles two exports you’ve been merging by hand every month. A one-page calculator your customers can use on your website. A document that turns your intake notes into a formatted proposal. Small, specific things that no software company will ever build because your version of the problem is yours alone.
Who Should Pay Attention
Anyone who has ever said the words “there should be a tool that just does this.” There now is, and you make it.
It matters most if you’re spending real hours on work that’s repetitive but slightly different every time — the kind that resists ordinary automation because it needs a small judgement at each step. That’s the category these tools are good at and templates never were.
Two honest limits. Both of these sit on paid subscriptions, so there’s a monthly cost before you build anything. And the Anthropic figures come from organizations already using the tool, which means office knowledge workers more than solo owners. The behavior is real; whether it looks the same in a two-person clinic is an inference on my part.
The Owner’s Forecast
Over the next year, the phrase “we need custom software” stops meaning what it has meant for thirty years. Not for large systems, which will still take real engineers, but for the small internal tools that make up most of what a small business actually wishes it had.
What I expect to follow is a quieter change in who gets hired. Some of the work that used to go to a freelancer as a small build now doesn’t leave the building. That’s good news if you’re the owner and worth planning around if part of your income comes from doing those builds for other people.
High confidence on this one, because it isn’t a prediction about adoption. The tools shipped, the usage data is published, and the direction of it is already visible in a sample of 600,000 organizations.
The part I’d watch is the skill rather than the software. Getting a useful result depends almost entirely on how precisely you can describe what you want, which is why two owners with identical subscriptions will get very different value out of them.
Your Next Move
Good — 20 minutes. Pick the most annoying repetitive task in your week and describe it out loud, start to finish, as though you were explaining it to a new hire. Type that description into whichever assistant you already pay for and ask it to build you something that handles it. You’re not committing to anything. You’re finding out what the current answer looks like.
Better — an afternoon. Take one thing you’ve wanted for years and never justified paying for. The calculator, the checklist generator, the report that pulls two things together. Build it, use it for a week, and change the parts that annoy you. The changing is the part that used to be expensive and now costs nothing.
Best — ongoing. Start keeping a list. Every time you catch yourself doing something a machine should be doing, write one line about it and move on. Once a month, take the list and build the one at the top. That habit is worth more than any single tool on it.
Today’s Instrument
Whichever one you’re already paying for.
Claude Cowork and ChatGPT Work are close enough for this purpose that the right choice is the subscription you already have. We have no commercial relationship with either company and earn nothing from today’s issue. If you have neither, both offer a way to try before paying, and one month is enough to know whether this fits how you work.
Tomorrow: what separates a description that produces something useful from one that produces a mess.
Owners Are Asking
Do I need to know anything technical to try this?
No. The whole change is that describing the thing has replaced building the thing. If you can brief a contractor clearly, you can do this. The failures come from vague descriptions, not from missing technical knowledge.
Is what it builds actually usable, or is it a toy?
It varies, and the variation tracks how specifically you asked. Small internal tools, calculators, formatted documents and spreadsheets come out genuinely usable. Anything customer-facing that handles payments or personal data deserves a second pair of eyes before you rely on it.
Which one should I pick?
For most owners it isn’t worth agonizing over. They overlap heavily. The rough difference reported so far is that Cowork is stronger with files sitting on your own computer, and ChatGPT Work is built more around connected cloud apps. Start with whichever you already pay for.
Does this replace the software I already run?
Not the platforms doing heavy lifting — your booking system, your accounting, your payments. This fills the space between them, which is where most owners lose their hours.
Sources: Anthropic, Claude Cowork usage analysis (1.2 million sessions across 600,000+ organizations, May 2026) · Anthropic product announcement, July 7, 2026 · OpenAI, ChatGPT Work launch, July 9, 2026