Anthropic finance agents show vertical SaaS is just agents

Anusheel BhushanAnusheel Bhushanco-founder of PlankOriginally posted on LinkedIn 

Anthropic releases agents for various financial services and insurance use cases. Most of vertical SaaS, at its essence, is these agents. Anthropic is making Claude the interface for knowledge workers and different use cases to just be agents created by domain experts. The agents are connected with underlying systems of record. What is missing in this architecture?

anthropic.com/news/finance-agents

table
  • Enterprise AI
  • AI agents

Keep reading

Software architectureAI agents

Absurd, a Postgres-native workflow engine for edge agents

Really cool Postgres-native workflow system. The agent's entire pause/resume state lives in the DB; the worker process can exit. For running AI agents on the edge on end client machines, this is the unlock: because the whole engine is one SQL file + stored procedures with no broker, no coordinator, no SDK-based orchestration in code. It also makes the entire execution workflow history immutable and viewable by a self-reflection agent. On local/edge where processes die routinely (laptop sleeps, browser tab closes, network drops, edge worker suspends), an agent that has already made 20 expensive tool calls resumes at step 21 instead of re-paying for everything. You get this without writing checkpoint code or a state machine; the trace of completed steps is the agent's memory. I’m a fan of the pattern where the data itself captures the state in an idempotent sort of way. earendil-works.github.io/absurd

AI agentsSoftware architecture

Separate working agents from codegen agents that build them

As agent prompts become an important part of the codebase, there needs to be an explicit separation of an agent with a prompt that does something and a codegen agent that generates another agent. An example is in scraping, a first agent uses the screenshot and DOM structure to fetch some data. And a meta codegen agent generates a piece of typescript code + prompt that is essentially the first agent. So the codegen agent is creating the prompt and agent graph the agent that does the work.

Enterprise AI

Enterprise knowledge is the new walled garden in AI startups

A new type of walled garden is now taking shape with enterprise knowledge, as they translate their workflows into closed-source AI startups. Both the code, as well as data, are going into these new silos. Enterprises are spending a bunch of effort to translate their workflows, processes, and memory into proprietary, closed source, non-interoperable structures that live in the source code and databases of some short-lived AI startup. This needs to be addressed with a new type of open intelligence system similar to open source. It requires rethinking the deployment, licensing and ownership of code, data and memory.

Coding agents

The product trap of harness engineering and agentic coding

Harness engineering and agentic programming are super addictive and fun, there's strong parallels with video gaming. However, this will lead to a lot of smart, non-technical folks building software just for the sake of it without clarity on user pain points. I call this the “product trap.” Building software is gratifying in the minutes and the hours people are building small tools for themselves, but the act of creating the software itself is not a substitute for solving real user pain points

All postsOlder posts