Hype vs. Reality: Where AI Fits in the World of MGAs

The headlines say AI will revolutionise underwriting, automate placement, and optimise portfolios in real time. Providers are hyping the scope and now encompassing the broad generalised AI concept “Artificial General Intelligence” (AGI), including automated portfolio underwriting and ‘command centres’ guiding the agents to automate ‘thousands’ of underwriting use cases. But anyone who’s lived through the internet boom or the SaaS revolution knows — transformation never happens by magic.

I remember being in San Francisco in the early dot-com years, when a consulting firm pitched themselves as “redefining consulting for the internet age.” Their opening funding pitch slide read:

“I eat raw meat, and I love the kill.”

That was the culture — speed, sales, and bravado. My boss made a billion before the bubble burst.

Later, a marketing lead at one of the world’s largest software companies — still a major investor in AI today — taught me another lesson. I was preparing a product pitch when he stopped me and said: “Just write down what the customer wants. That’s what the product does.”

It was meant as a joke, but it captured something real: tech has always been better at selling possibility than proving capability.

And that’s where the insurance world — and underwriting in particular — draws a hard line. Because underwriting is built on trust and evidence, not hype.

The AI Layer and the Database Debate

There’s a growing view in tech — popularised by Satya Nadella — that all enterprise applications are really just databases, and that the logic is moving into AI.

If that proves true, it could render much of the industry’s recent transformation effort — browser-based PAS systems, cloud hosting, SaaS re-skins — transitional at best.

In this new model, the AI becomes the logic layer. It reasons across knowledge graphs that link both numbers and words — structured and unstructured intelligence.

That’s powerful for insurance.

Imagine an AI that can connect claims ratios with loss narratives, or combine pricing performance with broker feedback.

But let’s be realistic. We’ve seen this movie before. The internet took ten years to mature from dial-up pages to mobile platforms. AI will take maybe, half a decade, of infrastructure building — datacentres, regulation, and trust — before it reaches that level of maturity.

And the moment you can’t explain what’s happening inside the “magic”, trust evaporates.

How AI Actually Fits in the MGA World

Let’s bring it back to our world — to entrepreneurial MGAs. They’re often founded by brilliant underwriters with a track record and limited resources. They don’t have integration armies. They need tools that ‘work out of the box’, but still scale, end to end, with ambition.

It’s the skinny vs fat PAS debate. For carriers, skinny PAS makes perfect sense. They often run multiple systems — one per region, per line, or per legal entity — and centralise key functions like pricing, rating, and portfolio management. Skinny systems are cheaper to deploy, easier to swap, and allow flexibility across large portfolios.

For MGAs, it’s a completely different story. MGAs don’t have the scale or integration teams to weave a dozen systems together. They need a single system, ideally, that is richer, configurable, multi-class, and capable of doing the heavy lifting end-to-end.

Narrow AI or AGI in an MGA?

Underwriting in carriers seems to be starting with expensive human processes, the heavy rekeying with ‘quote loading’ offshore teams or worse, expensive underwriters. AI companies like MEA, Cytora and FurtherAI have great products. Easy to integrate to and providing immediate cost savings. They have figured out transparency and repeatability.

Providers are now extending the predictive analytics and pricing tools to encompass LLMs. It’s the PAS debate again, but it will take time to bring AI into automation of the logic layer.

MGAs ideally want pre-integrated AI tools and a single system for all classes. That’s why we build MGA Connect on Salesforce. MGA Connect can easily leverage best of breed AI tools or leverage Agentforce. It always takes time for the horizontal tools like Salesforce and Agentforce to catch up and commoditise the specialist insurance technology providers. It is worth remembering that all the AI providers are leveraging the same Anthropic, Google and OpenAI LLMs, which may challenge the vertical niche vs the best of breed horizontal tools.

Over time, Agentforce’s AI agents can handle submission ingestion, summarisation, triage, and enrichment — while MGA Connect manages the end-to-end policy workflow, audit trail, and compliance backbone. Together, they allow underwriters to handle more business, faster — without losing the human judgement at the heart of underwriting.

It doesn’t promise a “self-driving MGA.” It delivers measurable gains in speed, accuracy, and transparency.

And because it’s built on Salesforce, it’s ecosystem-ready. You can plug in best-in-class tools — pricing engines, analytics platforms, document AI — without custom builds or brittle integrations.

It’s fat where it counts, open where it matters, and smart where it helps.

Trust, Testing, and Repeatability

In the end, every conversation about AI in insurance comes back to one thing: trust.

Regulators, capacity providers, and brokers all demand evidence — not anecdotes.

That’s why we’re making testing, transparency, and repeatability central principles of how we build.

Agentforce’s AI outputs are explainable and auditable. Every transaction within MGA Connect can be traced. If the AI flags a risk or suggests an action, you can see why.

That matters, because underwriting isn’t a black-box business. It’s a trust-based one.

And while AI will become more sophisticated — blending portfolio analytics, LLMs, and pricing data — the human dynamic between broker and underwriter won’t disappear. Technology can support it, but it can’t replace it.

We don’t believe in magic.

We believe in evidence, ecosystems, and enabling the underwriter.

That’s how you make AI work for the real world of underwriting.