Caddi

There is a specific kind of work that resists software. Not the glamorous work - the intake form, the document that has to move from one system into another, the invoice that needs to match a time entry, the calendar that never syncs. Professionals at law firms and financial advisories spend an alarming share of their day doing this, and every attempt to fix it has historically asked them to become, temporarily, a little bit of an engineer. Learn the automation tool. Map the fields. Build the flow. Most people, reasonably, decline.

Caddi's premise is that this is backwards. Instead of teaching the expert to build automation, you let the expert do what they already do - the job - while the software watches. A person records their screen, narrates what they're doing the way they'd train a new hire, and Caddi's system extracts the intent behind each click. Then, and this is the part that matters, it generates actual code that talks to the underlying application APIs. Not a bot clicking pixels. Not an AI agent improvising at runtime. Deterministic code, the same way twice.

The distinction sounds academic until you remember who the customers are. A law firm cannot have an automation that is usually right. Neither can a wealth manager or a mortgage company. "Predictable" is not a nice-to-have in regulated professional services - it is the entire ballgame. Caddi built its whole thesis around that constraint, which is a more interesting bet than it first appears.

Plenty of companies promise AI that automates your work. Caddi's twist is philosophical, and it shows up in the engineering. Most AI automation is autonomous - an agent decides, at runtime, how to handle each situation. That's flexible and occasionally impressive, and it is exactly the wrong property for a firm where a wrong answer has legal consequences. Caddi instead generates fixed code from a demonstration, trading some flexibility for the thing regulated customers actually want: the same result, every time, that a person can inspect.

There's a second, quieter benefit. When you record how an experienced staffer does a job, that knowledge stops being locked in one person's head. If they leave, the workflow stays. For a 20-person firm, that's less a feature than a form of insurance - and it's the kind of thing that doesn't fit on a pricing page but tends to close deals.