Hyperexponential reports 50% faster submission-to-quote times and says it supports 45bn in GWP across more than 40 enterprise customers, including Aviva, Beazley and Markel.
Amrit Santhirasenan argues capabilities that were a whole product six months ago are now available within Gemini or GPT-5, so insurtech moats must shift toward proprietary data and workflow.
Santhirasenan says the junior underwriting job market's collapse is an existential challenge, and he urges leaders to optimise AI adoption for change rather than immediate results.
Why insurance underwriting is becoming a human and agent partnership
Underwriting is moving to a model where swarms of AI agents handle intake, triage, pricing and research alongside human underwriters and actuaries. Amrit Santhirasenan argues the edge now lies in the information fed to decisions rather than the models, since frontier capabilities commoditise within months. Faster quotes and better loss ratios follow, and better pricing lets the world take on more risk.
Takeaways
Better information beats better models. Amrit argues GLMs, long the industry standard, are no longer state of the art, and the real gap is the data reaching decision makers. Deep research on every risk now costs about $20 a month, making richer inputs affordable at scale.
Agent swarms compress quotes from weeks to hours. Hyperexponential runs ingestion, actuarial and underwriting agents alongside human staff and is adding a triage offering. Generative AI flags risks red, amber or green far faster. Quotes that took about two weeks return within hours, and Hx reports 50% faster submission-to-quote.
Freed underwriter time is a strategic choice. McKinsey research cited by Amrit points to 3 to 6 percent loss-ratio improvements, of which Hyperexponential takes a conservative slice. Where the surplus from saved time goes is up to each company, though he wants underwriters spending more time on borderline risks.
Junior job collapse threatens the talent pipeline. Amrit calls the disappearing junior underwriting market existential for insurance and the wider economy, though AI can explain algorithms to new staff. Technical skills commoditise faster than relationship skills, yet Jevons paradox and a protection gap near $2tn imply more aggregate work.
Compliance means labeling and auditing agents. Agent opinions must be unambiguously identified as non-human. Bias cannot be perfectly solved but can be audited, possibly by a separate judgment and bias agent supervising the rest. Explainability pressure is rising in US admitted markets, while AI-native carriers may tolerate more agent risk.
How carriers should adopt underwriting agents
Make adoption leadership-led. Amrit points to Paul Brand setting up the Expensify system himself.
Optimise for change rather than results, and measure progress along the way.
Start with jobs to be done and pilot in reversible two-way-door areas, moving through them quickly.
Avoid rip-and-replace of core systems such as ledgers.
What Amrit expects next
Much better service within months, with specialty quotes taking hours.
Tap a name to go there · tap a line to see why they are linked
On this episode 2
Show 1
New hybrid jobs, and technical underwriters finding actuarial work easier than ever.
Brokers, already in active Hx pilots, becoming among the most invested in algorithmic underwriting.
Every Hyperexponential client using its agents at least monthly by the end of next year.
Better pricing letting the world take more risk and shrinking the coverage gap.