When Anthropic shipped Sonnet 5.5, the demos that made the rounds were games. Crownfall. Bounce Lab. Beautiful, technically impressive, and completely irrelevant to a shop that fixes HVAC units or ships custom cabinetry.
But underneath the toy demos is a milestone worth paying attention to: the mid-tier models — the cheap, workaday tier — can now build working software in a single sitting. Not frontier-priced. Not developer-gated. The tier you already pay for as part of a normal subscription.
For a 10-person business, that changes a specific calculation: the internal tools you've been putting off because "we'd need a developer for that" are now buildable without a developer.
Why "Internal Tools" Is the Right First Target
Small businesses don't lack software. They lack software that matches how they actually work, so they make do with spreadsheets, whiteboards, and tribal knowledge.
Internal tools are the ideal first build for three reasons:
- Nobody's job depends on it. If the tool breaks, the work continues the old way. There's no customer staring at an error page.
- You're the only user. You can change the spec, the fields, and the workflow without a migration project or a change-management meeting.
- It builds the skill that transfers. Describing a tool precisely enough that an AI can build it is the same skill as describing a process precisely enough that an AI can run it. Internal tools are the practice ground.
What the Workflow Actually Looks Like
This is not "type a prompt, receive software." It's a tight loop:
- Write the spec in plain language. What gets tracked, what states a job moves through, who updates what, what the screen should show. One page. If you can't write it in plain language, the tool isn't ready to exist.
- Let the model build the first version. Expect rough edges. The goal is something you can click through, not something you can demo.
- Use it yourself for one week. Keep a note open and record every annoyance — every field you wish existed, every click that felt dumb.
- Feed the list back and iterate. Small fixes, described specifically. This is where the tool becomes yours.
- Keep the data exportable. CSV export, a backing spreadsheet, something. The tool is disposable; the data is not.
The first version takes an afternoon. The second version takes another afternoon a week later. That's the whole schedule.
The First Three Tools to Build
If you're wondering where to start, build in this order — from lowest risk to highest:
1. A Job Tracker
Every service business runs work through informal stages: quoted, scheduled, in progress, done, invoiced. Off-the-shelf tools force your stages to match their flow. A model-built tracker matches yours, because you describe yours. Start with: job name, customer, stage, next action, who owns it, date. Resist adding anything else for two weeks.
2. A Quote Generator
Your material costs, your labor rates, your markup logic, your standard options — in one place, producing a consistent quote every time. This is the tool that quietly removes "let me check with the owner" from standard quotes. The owner's judgment gets encoded once instead of being pinged forty times a week.
3. A Weekly Numbers Dashboard
Point it at your accounting export or your master spreadsheet and have it build the view you actually want: revenue this week, quotes outstanding, jobs in progress, invoices due. The value isn't the chart — it's that you look at the same four numbers every Monday without assembling them by hand.
The Guardrails (Read This Part Twice)
Model-built tools are prototype-grade. That's a feature when you're starting; it's a liability the moment something important depends on them.
- Keep customer data out until you've read the terms. Know what your AI subscription does with data you paste in, and don't route payment details or sensitive customer information through a prototype.
- Don't connect it to money. A dashboard that reads your numbers is fine. A tool that moves them needs real access controls, backups, and a human who owns it.
- No customer-facing tools yet. Internal first, always. Customer-facing software has a failure cost that internal tools don't.
- When a tool becomes load-bearing, harden it deliberately. If the whole team now runs on the job tracker, that's your signal to add access control, backups, and a support plan — or rebuild the same tool on a supported platform. Promotion from prototype is a decision, not a drift.
The Demo That Matters
The Sonnet 5.5 videos with the view counts are games. The demo that matters for your business is a boring job tracker that your team actually opens every morning — built in an afternoon, fixed over a weekend, and owned entirely by you.
Start with the tool nobody's job depends on. Keep the data portable. Treat every build as a draft until it earns promotion.
Want the workflow templates and review checklists for your first AI-built tools? The AI Automation Starter Kit includes spec-writing frameworks, iteration prompts, and the guardrail checklists to keep prototype tools from becoming liabilities.
Top comments (0)