Our 2025 Year-End Perspective: “Every new beginning comes from some other beginning’s end”
Ian Levinsky
Closing Time - Supersonic (1998)
Part mixtape, part field notes: making sense of the year behind us and the one ahead by pairing song lyrics with lessons from building AI tools at Pathwork.
We feel pretty privileged and very lucky.
We started Pathwork in 2024 when AI-native was a newly minted phrase, and we chose an industry (Life, Health, and Annuities insurance distribution) that actually made sense to innovate. Big Market. High stakes. Wrapped in regulation. Deeply human workflows. Lots of paper cuts.
We’re writing this article to share what we experienced, heard, debated, and now believe for the four audiences we spend the most time with: distributors, carriers, investors, and software founders.
From our seat (and what we consistently heard from others), 2025 became the year companies stopped prototyping, POC’ing, and MVP’ing…and started productionalizing. Usage went up and expectations hardened. Tolerance for “cool demos” went down.
But, productionalizing only happened if you got the use cases right.
The teams that won built systems that were explicitly in-the-loop. UI-driven workflows grounded in partial autonomy and clear human checkpoints. The pragmatic winners looked more like Cursor: fast, assistive, opinionated, and useful. The failures felt more like AI travel agents that spend two hours booking you a car in Lincoln, NH when you needed one in Lincoln, NE.
Both are applications using the same class of foundational models but with very different outcomes.
Let’s dive in…
Distributors: “Done, done, on to the next one”
All My Life – Foo Fighters (2002)
Distributors, like agencies & BGAs, want to get more done, faster. Plain and simple. AI didn’t create that need. It made it clear the industry was finally ready to act on it. How the field should think moving forward, there are two trends we want to highlight:
First: the workforce cliff. Roughly 50% of the insurance workforce is expected to retire over the next 15 years, creating an estimated 400,000 open roles across distribution and operations (Source: LIMRA and ACLI).
Second: buyer behavior has already changed. Nearly 6 in 10 young adults say they would use AI to research life insurance, even if they still prefer to buy with a human advisor (source: LIMRA). Broader AI adoption is accelerating fast: over 50% of adults ages 18–64 reported using AI tools in 2025, up sharply year over year (Source: Federal Reserve Bank).
Distribution implication: Applicants and agents now show up in AI-research mode.They expect: 1) Instant comparisons 2) Plain-English explanations 3) Clear next steps. All before they ever talk to an advisor.
So for distributors that want to win:
Meet prospects where they already are (AI-powered education, instant answers, pre-screening)
Maintain a clean, intentional handoff to humans for confirmation, trust, and close
Bain & Company has repeatedly highlighted operational excellence—not product differentiation—as the biggest driver of distributor outperformance. AI’s real value showed up in summarization, triage, and structured data capture.
A word of caution heading into 2026: Be skeptical of tools that promise to “use your computer” or fully navigate screens end-to-end. They’re expensive, slow, and error-prone today. The ROI just isn’t there yet from our experience.
Insurance Carriers: “I still haven’t found what I’m looking for”
I Still Haven’t Found What I’m Looking For - U2 (1987)
If I had written this section last year, the title probably would’ve been from Wilson Phillips’ “Hold On for One More Day.” We sensed an impending shift back then. In 2025, it showed up clearly.
Across the industry:
81% of insurance CEOs now rank generative AI as a top investment priority
90% say AI is their most important strategic initiative for 2026 (Industry CEO surveys, 2025)
Global spending followed suit. Gartner estimates GenAI spend reached $644B in 2025, up 76% YoY.
Two patterns stood out:
1. Human-in-the-loop is non-negotiable. Carriers remain deeply uncomfortable with LLMs making final decisions autonomously. And frankly, it’s a good instinct.
2. AI is being pushed into distribution, not left as optional tooling. Multiple 2025 surveys showed:
More production deployments
Larger committed budgets for 2026
Rising usage inside core portals and CRM
Looking ahead to 2026, expect carriers to use AI in high-volume, low-margin workflows, where automation can materially improve unit economics. The other area we’re seeing innovation is around workflow enablers for wholesalers (systems that answer a very practical question): “Which producer should I call today, and with what product?”
Venture Investors: “Same as it ever was”
Once in a Lifetime – Talking Heads (1980)
AI code-gen absolutely compressed build times in 2025. At Pathwork, we experienced and measured it. What it did not remove (at all) was the need for product taste nor distribution risk.
Demand for vertical software keeps expanding, especially in regulated, operationally complex industries. Buyer attention, however, remains stubbornly finite. Decision cycles are still long. Trust still compounds slowly and distribution is still earned.
VC-backed roll-ups are facing increased skepticism as thin integration proves insufficient, competition for strong assets intensifies, and the burden of running M&A alongside a technology business becomes harder to sustain. Likewise, “AI wrappers” without proprietary data, real workflow depth, or structural embedding into how work actually gets done are starting to feel fragile.
At the same time, the pattern of what actually works looks familiar (thankfully) in a good way. Find great teams building in great markets with durable business models with a “why now” platform shift at their backs.
Insurtech and Vertical SaaS Founders: “Staring at the blank page before you…”
Unwritten – Natasha Bedingfield (2004)
…Open up the dirty window! If you’re building in insurtech, or any vertical SaaS category, this is an unusually good moment to start.
Not because the work is easy, but because the contours of real opportunity are finally visible. The use cases that matter have emerged. Voice AI works. Workflow automation works. Intake, summarization, and recommendation systems are no longer experiments. Buyers have lived with enough AI to evaluate it seriously… not as a novelty, but as serious enterprise infrastructure.
If you know a domain deeply, have a clear point of view on what’s broken, think legacy providers aren’t keeping pace and believe customers will care when it’s fixed, you should build. The new standard is actually useful. The teams that pull ahead may not be the most ambitious on paper, but they’ll probably be operators with good taste, solid problem-solving skills, commitment, and empathy. If you’ve been waiting for clarity, it’s here.
Conclusion: You don’t have to go home, but you can’t stay here
Closing Time - Supersonic (1998)
That wraps up our thinking for the year! Just kidding, we’re always thinking. If 2025 made one thing obvious: well-built, customer-centric AI tools are being used, every day. For 2026 the question shifts from “Can we use AI in our enterprise?” to “Where do speed, clarity, and trust create the most value?”
We hope you enjoy the end of the year with loved ones and wish you happy holidays, a happy new year, and cheers to a prosperous 2026!
Enjoying the climb,
Ian
Sources used:
https://www.wsj.com/tech/ai/ceos-to-keep-spending-on-ai-despite-spotty-returns-2eaeb6b9
https://content.naic.org/sites/default/files/call_materials/Model%20Bulletin%2010.23%20Clean.pdf?
https://www.ft.com/content/abfe9741-f438-4ed6-a673-075ec177dc62?
https://www.ijfmr.com/papers/2024/6/33609.pdf
https://insurancenewsnet.com/innarticle/how-ai-is-lowering-the-barrier-for-annuities-sales
https://kpmg.com/us/en/articles/2024/impact-artificial-intelligence-insurance-industry.html