AI & Automation

No, and the shortage is not technical. Plenty of people can call a model; far fewer understand a specific business well enough to know which repetitive task is worth removing. The durable position is domain knowledge plus delivery, not model access. If your only advantage is knowing the tool exists, that advantage disappears the moment your customer also knows.

Most of them can, and most of them will not. What people pay for is the thing already set up, already reliable, already fitted to their process, with someone accountable when it breaks. Sell the finished outcome and the responsibility for it. If your product is genuinely just a prompt, expect it to be replaced.

Assume both will happen. Anything built as a thin layer over a single model is exposed to its pricing, its policies and its next release. What survives is the workflow around it: your customer data, their process, the integrations, the trust. Build so that a model swap is an inconvenience rather than an extinction event.

Price the result, not the build time. The customer is buying hours returned or mistakes avoided, and neither is cheaper because your tooling is good. Explaining how simple it was is the fastest way to devalue it. If ten minutes of your work saves them a day every week, that is what it is worth.

Too late to compete on model building, almost certainly. Not too late to apply it to a specific industry that has barely begun. The unglamorous, domain-specific problems are the ones still open, and they are open precisely because they require knowing a business rather than knowing a model.

Less than you probably assume, and the consequences of getting it wrong are legal rather than technical. Customer data belongs to your customer, scraped data may carry terms you have agreed to, and personal data carries obligations regardless of how you obtained it. Decide what you will and will not touch before you build, and get advice if you handle anything sensitive.