Will the big labs leave anything for external competition?
This probably killed a thousand startups in this space.
in the early internet you wouldn't see google creating their own news site or facebook building their own animal farm.
what happened to platformication of everything?
Or "rent seeking" type tools/options that should be baked into products from a much earlier point.
There was an app for OSX that added window snapping, long before Apple added it to their desktop environment. $5 or something for a free feature that just makes sense to build into your product from the start. Apple is king at absorbing this sort of paid add on, eventually. AI makes that faster.
Yes, in very specific cases where I fully understand the methodology(ies) that is (are) applicable, and am able to verify correct implementation. Also, as an enhanced ‘Google search’ to supplement what I have found. I am the skeptical type… yet, so far have been impressed. But, I wouldn’t trust using AI to blindly give me solutions to a problem I couldn’t solve myself, albeit much more slowly.
I don't trust these AI-only companies to be overnight experts in properly handling medical, financial and insurance data. They have no business providing these tools, unless they want to take all the risk too.
Making the most convoluted and idiotic insurance process on earth and then delegating that process onto an AI that requires huge buzzing data centers.. Is there an option to respawn in the non-clown world universe? It was funny at first but it gets tiring eventually.
Why would this be useful in a zero sum environment like markets, why would you want to use the same tool that everyone else has access too? Top performers will always be the people that hand craft their solutions, just like why the top performers in the watch space are the people that make handmade watches in Switzerland not the guys make 100k watches a month in China.
"ready-to-run agent templates for the most time-consuming work in financial services: building pitchbooks, screening KYC files, and closing the books at month-end"
Ok, maybe you can squeeze a vaguely passable pitchbook out of Claude.
But screening KYC files or closing books at month-end ?
"I'll have some of what they're smoking" as the cool kids say.
No regulator or tax office on this planet is going to accept the "but Claude said it was ok" excuse.
The only people who are going to profit out of this are Anthropic, Lawyers and Governments (through increased fines).
Great to see more insurance hype! We've been working on AI to solve the consumer search problem in the industry for the past 3 (almost 4) years and it's great to see the big labs getting their hands dirty and building tools for practitioners in the space.
More industry exposure to well-managed agentic experiences will create oodles of opportunities to reduce premiums for consumers and offput some inflation-driven increases in cost of coverage.
Does anyone else think "agents" are the wrong abstractions? Agents look like UI wrappers over LLM's - they are inherently not composable. Tailor made agents for UI's don't seem to scale. I predict they wont take off.
What I predict instead is that we will have a common UI layer plugin and a "protocol" than can speak to ui elements -- this might be more composable.
Of course - finance is the best domain to depkiy a stochastic parrot which hallucinates and forgets stuff frequently and doesn’t follow your instructions - even with SOTA models. One where you need absolute accuracy and auditabikity.
This is great but as someone in infrastructure tech at a large financial, there is almost no framework for cleanly separating control from data plane operations, read vs write, anything. As of right now you have to build nearly all of that yourself.
It feels like juggling pipe bombs and I have a ton of empathy for the teams being pressured by the business to roll them out with no appreciation for the regulatory rat's nest that ensues.
I've been doing bias and misaligned behavior research, creating custom private eval suites to test and compare models. Claude Opus 4.7 is heavily biased and presents clear regulatory and reputational risk.
It seems the initial product footprint tries to sidestep this problem by not giving the agents control on who to lend to or which applications to approve. Even so I think it's quite an optimistic read on their end. Happy to share reports to anyone who's interested (montana@latentevals.com), especially if you work at a frontier model lab and are interested in plugging my evals into your RL systems!
Nobody is using LLMs to make lending decisions. They are using LLMs to read, extract and audit the supporting documents that go into normal well-tested, compliant and rules-based underwriting systems. And firms A/B test against humans doing the same work. The outcomes your are looking for are metrics like delivering faster results back to customers, with fewer mistakes and less fraud, more compliant, than a comparable human-only process.
39 comments
[ 3.5 ms ] story [ 44.5 ms ] threadThis probably killed a thousand startups in this space.
in the early internet you wouldn't see google creating their own news site or facebook building their own animal farm. what happened to platformication of everything?
There was an app for OSX that added window snapping, long before Apple added it to their desktop environment. $5 or something for a free feature that just makes sense to build into your product from the start. Apple is king at absorbing this sort of paid add on, eventually. AI makes that faster.
I've really only seen it used for research / exploration thus far. Either for economic research slide deck or for exploring trading hypothesis
The templates being: pitch builder, meeting preparer, earnings reviewer, model builder, market researcher, valuation reviewer, general ledger reconciler, month-end closer, statement auditor, KYC (Know Your Customer) screener.
Seems pretty scattershot. Reminds me of GPT Store.
Is the plan to have an LLM do everything? And do it worse?
"Oh yeah my Claude didn't agree with the pitch from their Claude"
The goal of current tech is to make humanity a gerbil running on a Claude wheel
Better Call Saul when (not if) it does.
"ready-to-run agent templates for the most time-consuming work in financial services: building pitchbooks, screening KYC files, and closing the books at month-end"
Ok, maybe you can squeeze a vaguely passable pitchbook out of Claude.
But screening KYC files or closing books at month-end ?
"I'll have some of what they're smoking" as the cool kids say.
No regulator or tax office on this planet is going to accept the "but Claude said it was ok" excuse.
The only people who are going to profit out of this are Anthropic, Lawyers and Governments (through increased fines).
More industry exposure to well-managed agentic experiences will create oodles of opportunities to reduce premiums for consumers and offput some inflation-driven increases in cost of coverage.
As someone who has been interviewing lately, I think this is the next step after leetcode and whiteboard style interviews.
What I predict instead is that we will have a common UI layer plugin and a "protocol" than can speak to ui elements -- this might be more composable.
Why didn’t I think of that.
however the result (excel/spreadsheet) looks different each time you run it. Which is annoying when you run it at the end of each month.
btw: this is not surprising when you look at the low details the skills have.
It feels like juggling pipe bombs and I have a ton of empathy for the teams being pressured by the business to roll them out with no appreciation for the regulatory rat's nest that ensues.
It seems the initial product footprint tries to sidestep this problem by not giving the agents control on who to lend to or which applications to approve. Even so I think it's quite an optimistic read on their end. Happy to share reports to anyone who's interested (montana@latentevals.com), especially if you work at a frontier model lab and are interested in plugging my evals into your RL systems!