SEATTLE, Aug. 24, 2026 /PRNewswire/ – Caddi today launched an AI agent that builds and governs a firm's AI agents. It is aimed at the industries that run the largest back offices: wealth management firms, law firms and insurance companies.
The agent finds the repetitive work already running inside a firm's own systems and ranks it by what it costs. It is trained on each process the way a new hire would be, working through the edge cases and failure modes with the person who knows them until the process is right. Every agent it builds then executes deterministically, and every run is logged against a scoped permission.
The company's argument is that the back office of a wealth management firm, a law firm or an insurance company was never left manual because the work was too boring to bother with. It was left manual because it is genuinely hard to specify, and every previous generation of automation software required someone to specify it perfectly in advance.
The work looks simple. It is not.
Filing an executed contract sounds like one step. In practice it is a decision tree. Did both parties sign, or only one. Is this the executed copy or another draft in the same thread. Is there already a version filed against this matter, and does this one replace it or sit beside it. Did the document arrive as a scan with no text layer. Are there two counterparties on this agreement rather than one. Is the effective date missing, and is that a reason to stop or a reason to proceed.
Opening a client account is the same shape. So is reading a stack of broker-dealer transition forms and getting them into a CRM, reconciling a day of client payments, running a conflicts check, or chasing the records a file is missing. The happy path is a handful of steps. The real process is the happy path plus forty exceptions, most of which live only in the head of the person who has been doing the job for six years and has never written any of it down.
That is the reason automating this work has historically cost so much. The detail was the entire project, and the detail was only discoverable by doing the job. Firms paid consultants to sit and watch, produced a specification, built against it, and then discovered the exceptions in production one incident at a time.
“Everyone in this market describes back-office work as low-value busywork, and that framing is why it never got fixed. It is not busywork. It is a process with fifty branches that one person carries in their head. That is a hard thing to automate and an easy thing to underestimate, and every tool that underestimated it shipped something that broke in week three.”Alejandro Castellano, co-founder and CEO of Caddi
Most firms cannot see the work in the first place
Before a firm can automate anything it has to know what it does, and at an organizational level most firms do not. Chief operating officers can describe their departments and name their systems, but very few can say with confidence which processes run across the firm, how often, who performs them, which tools each one crosses, or what any of it costs in aggregate. They have instincts about priorities. They do not have a source of truth, and so the first question of any automation program, where do we begin, gets answered by whoever complained most recently.
Caddi's agent starts there. It reads the systems a firm already runs and surfaces the processes that actually repeat, ranked by frequency, by how many tools each one hops between, and by what it is costing the firm. The output is a short list of named candidates with the evidence for each one attached. The first decision an operations leader makes is which process to automate, not whether automation is even possible, and it is made against observed activity rather than a hunch.
A proactive agent, not a passive recorder
All of this happens in Loop Studio, the canvas where a person and the agent design an automation together. Once a candidate is chosen, the person who does the job shares their screen, demonstrates the work, and talks through it the way they would train a new hire. The agent is not silently recording. It works the problem with them: it brainstorms how the process should be structured, offers best practices from processes of the same shape, and keeps asking until the ambiguity is gone.
The improvements it suggests are the firm's to take or decline, so an inefficiency does not simply get automated at higher speed. And whenever it encounters anything ambiguous it stops and asks, in the moment, while the person who knows the answer is still in the room. What if only one party has signed. There is already a version filed, do you overwrite it. Should wire confirmations be treated the same way as checks. What tells you an executed copy from a draft.
Each answer becomes a rule inside the agent rather than a note in a transcript. The interrogation is the point: it is a deliberate attempt to extract, in twenty minutes and while a human is present, the same body of exception handling that a traditional automation project discovers over six months of production incidents.
“The detail is the work. Every automation project that has ever gone badly went badly because a detail nobody thought to mention turned out to matter. So we built an agent whose job is to go looking for those details on purpose, and to go looking while the person who knows them is still sitting there. That is a fundamentally different posture than recording someone's clicks and hoping the recording was representative.” Aditya Sastry, co-founder and CTO of Caddi
Deterministic execution, because long agent runs are where accuracy goes
The agents Caddi produces are hybrid. They call AI where the work genuinely requires judgment, such as reading an unstructured document or classifying an inbound message, and they run deterministic code everywhere the work has to be exact. Improvements are proposed between runs and approved by a person, never applied mid-run.
That architecture is a direct response to how general-purpose AI agents fail. When a model decides every step of a long task, per-step reliability compounds against it: a step that is right 95 percent of the time is right roughly three quarters of the time across six steps, and back-office processes routinely run far longer than six steps. Per-run cost also scales with the number of model calls, which is precisely the wrong shape for work that runs thousands of times a month.
A Caddi agent runs the same way every time. The same input takes the same path, at flat cost, whether it runs once or ten thousand times.
No black box
Every agent runs against a scoped permission for each system it touches, read, create or send. Every run is a replayable record of what the agent decided, why, and which permission it used. Firm leadership sees every agent in one place: who owns it, what it is allowed to do, and everything it has done. Caddi is SOC 2 Type II compliant and connects to more than 100 business applications, including Salesforce, DocuSign, Outlook, iManage, NetDocuments and Clio.
For firms whose regulators, clients and insurers can ask what the software did and when, the audit trail is not a feature alongside the automation. It is the precondition for deploying any of it.
What is at stake for the industry
Professional services is the last large sector of the economy without an operations layer. Manufacturing got ERP. Sales got CRM. Software engineering got continuous integration. The operations floor of a law firm or an advisory practice got a person with a checklist and an institutional memory nobody else shares, and it still has one.
The largest firms are responding by hiring engineers to build this in house. That is the wrong path at any size: an engineering team is the most expensive way to solve an operations problem, and the specification it needs is exactly the thing nobody can write. Smaller firms are priced out of even making the mistake. Over the next decade of consolidation, the firms that get their operations automated without building a software team will absorb the firms that do not.
Proof already in production
Caddi runs in production across wealth management, law and specialty finance.
In wealth management, customers include a Barron's Top 10 registered investment advisor, which uses Caddi to read broker-dealer transition paperwork and populate Salesforce as it absorbs growth from 1,500 advisors toward 5,000 without expanding its transitions team; and The Planning Center, an independent RIA that has run Caddi for more than a year across seven workflows and nine tools that have no native integrations between them.
In law, customers include a high-volume firm, Palace Law, which absorbed a near tripling of daily inbound mail without adding a person; and an Am Law 100 firm that automated 150 conflict checks a day.
“We thought our processes were finely tuned, but every one still pointed back to a human. Caddi got the systems we already have to talk to each other, and gave my team back an hour a day.”Matt Mercer, chief operating officer of The Planning Center
What this launch means for Caddi
Caddi raised a $5 million seed round led by Ubiquity Ventures, with participation from Founders' Co-op and the AI2 Incubator, on a thesis the company called record to code: that a demonstration of a process could become working, API-driven automation. That thesis held. What it did not answer were the three questions that decide whether a firm adopts it. Which process should be automated first, and on what evidence. What happens the first time reality does not match the demonstration. And how anyone proves, afterward, what the software actually did.
The agent answers all three in one product. It finds the work and ranks it by impact. It goes after the exceptions on purpose, while a human is present to resolve them. And it reports on itself, run by run, in a record a compliance officer can read.
“Every launch in this category promises to democratize AI. The scarce thing was never the promise. It is a firm that will let the software touch a client file, and that is a trust problem, not a model problem. Trust here is made of two things: getting the details right before you go live, and being able to show your work afterward. We built for both.”Alejandro Castellano, co-founder and CEO of Caddi
Availability
The agent and Loop Studio are available today. Firms can watch the full product walkthrough and start a free 14-day trial at trycaddi.com.
About Caddi
Caddi builds AI agents for the back office of professional services firms. Founded by Alejandro Castellano and Aditya Sastry and backed by Ubiquity Ventures, Founders' Co-op and the AI2 Incubator, Caddi is used by wealth management firms, law firms, insurance companies and other back-office-heavy businesses to automate the operational work that keeps their teams from their clients. Caddi is SOC 2 Type II compliant and connects to more than 100 business applications. Learn more at trycaddi.com.
Media contact
Jason Alafgani
Head of Marketing, Caddi
jason@trycaddi.com
Do more with less
See Caddi in action
Tell us where to reach you and the calendar opens right here. In 30 minutes we'll show you how Caddi automates the back-office work that grows with your clients—built, run, and maintained for you.