What we're thinking about.

Stop bolting AI onto the process you already have, and redesign the work around it. It's advice I've given more than once, and I still think it's right — but as stated it doesn't tell anybody what to move. Factories hit the same wall with electricity: they swapped the steam engine for a motor and turned the same drive shaft, and productivity didn't respond for four decades. The change that mattered was pulling out the shaft, which had been quietly fixing the floor plan. In underwriting the shaft isn't the policy admin system — it's the book. A submission is assigned first and read second. Move the point where the human attaches, leave the authority letters and producer relationships exactly where they sit, and you might have the other thing.

Ask a carrier CEO where AI touches a decision about a policyholder, and you'll get the list of things the company bought as AI. That list is accurate — and incomplete, because some of the AI reaching your customers arrived bundled inside software bought for a different job: a fraud score inside the claims platform, a propensity model inside the retention tool. Each one shapes a rate, a coverage decision, or a claim, and none of them are on the list. You can't govern, explain, or answer for a decision path you haven't found — and someone is eventually going to ask you to produce the real list. The fix isn't a governance program; it's a mapping exercise you can run on one workflow, on your own clock, before someone else asks on theirs.

Naming the verbs in a workflow tells you what each step could point to — a rule, a model, a language model, or a person. It doesn't tell you how much room to give that step once you know, and that second question is a separate decision most teams skip. Rules are predictable going in and rigid once written; agents are flexible and only traceable after the fact. Most shops make that trade once, at the project level, so the vendor's default configuration ends up deciding it for them. The fix is smaller than it sounds: for each verb you've already named, ask whether you chose where it sits on that line, or whether something else chose it for you.

Most carrier AI conversations start with the wrong question. "Where should we use AI?" gets answered at the altitude of departments and product categories, where nothing operational ever happens. The useful question is smaller: what are the verbs? Any piece of work is a chain of verbs, and each one points at exactly one capability — a rule, a model trained on your own data, a language model, or a person. Name them and you can read any workflow in your shop yourself, in about the time it takes to read this post — and the verbs that point at a person are the half of the map a rented system can't touch.

The retirement wave is a deadline on your AI program, not an HR problem. What walks out with a thirty-year underwriter is judgment — the tells, the exception paths, and the accounts she treats differently for reasons she never said out loud — and judgment was never written down. The practical move is to sequence the AI program against the succession plan, starting with the workflows where the honest answer to "who else can do this" is one person's name.

Launchpad accepts rolling applications from MGAs. We fund and build your AI infrastructure, you own the code, and you pay only on results.

Carriers are right to rent most of the AI stack — nobody builds their own cloud. But the agentic workforce built around your book and your underwriters' judgment is the one layer you can't afford to rent, because that's where carrier-specific advantage compounds. Rent that from a vendor and you've rented your own moat back from a business whose model depends on you never owning it.

Why carrier AI pilots break on the cases that matter most — and the one step that prevents it. Every operation drifts from its documentation the moment it meets reality. Build automation against the paperwork and you automate the routine majority while breaking on the exception-dense minority, which is exactly where the loss dollars, the litigation exposure, and the senior judgment already live.