What If an AI Lab Ships a Product for Your Industry?
Vertical AI products are arriving fast — here's the last 20% small teams should still own.
The most important AI news for small teams this week was not a new model — it was a vertical. OpenAI launched ChatGPT for Financial Services, aimed squarely at the kind of work junior Wall Street bankers do. In a live demo, the product analyzed a potential M&A target and pulled financial figures from industry-standard data sources, and the company signaled that tailored products for other sectors are coming.
If ChatGPT Work, released in July, was an agent that handled documents and repetitive tasks, this next step layers the way a specific job actually gets done on top of it. September updates point the same way: Deep Research expanded into Work and Codex, and files in Box, Dropbox and SharePoint can be cited without re-uploading them. General tools are no longer producing answers — they are producing deliverables.
Why are vertical AI products arriving now?
Because that is where the money is. OpenAI CFO Sarah Friar told investors in August that the company's enterprise business brought in more revenue than its consumer business. As general chat quality flattens out, the next growth comes from producing job-specific artifacts: slide decks, financial models, filings.
For a small team, that reads two ways. First, if your core value was 'we write good prompts and generate documents,' that layer is about to become a default feature. Second, if you understand the messy reality of one industry — forms, regulations, approval chains, where the data actually lives — that part is still not on anyone else's roadmap.
What's left when a big lab enters your market?
Big tech takes the generic 80%. The last 20% is what customers actually pay for, and it usually comes from access, accountability and format rather than raw model quality. These five areas stay defensible for a small team.
- Data access: booking systems at regional clinics, public procurement portals, a client's internal ERP — places generic connectors don't reach
- Format and rules: local tax and payroll forms, industry filing templates, contract conventions where one wrong field means rejection
- Accountability: who checks the output and who is liable when it's wrong — approval steps, audit logs, review criteria
- The last mile: not just producing a document, but submitting, sending and registering it
- Relationships: speaking the industry's vocabulary and answering the phone in week three of rollout
Say you build tooling for small accounting firms. Extracting and summarizing line items from receipts is already handled well by general models. But mapping those items to the right ledger accounts for a filing format, flagging only the rows a human must confirm, and warning about missing documents three days before the deadline — that flow can only be designed by a team that has worked inside the industry. It is also usually the reason customers don't churn.
What should you do in the next two weeks?
Don't fight head-on; build on top. Split your features into generic and proprietary, hand the generic ones to big tools without sentiment, and concentrate your people and marketing on the proprietary side. Concretely:
- List your features and mark each one yes/no on 'could ChatGPT or Gemini get 80% of the way today?' Two hours is enough.
- For every yes, stop in-house development and replace it with an API or integration.
- Pick three 'no' features and build an eval set from 20 real customer cases. Build it once and reuse it every time a model changes.
- Rewrite your pricing around jobs processed or hours saved, not token usage.
- Change the first line of your sales deck from 'we generate documents with AI' to 'we complete your industry's filings to spec and review them for you.'
- Talk to customers first. On the day a big vertical product launches, silence makes them assume you're about to disappear.
Vertical AI isn't only bad news for small teams. The more someone else builds the generic layer, the less you have to build — features that once needed a whole team can now be assembled in days on top of it. But that shift only starts when you let go of the work everyone else can do. Pick one feature to drop this week.