LLM Cost Control & Internal Agent Operations
Wealth ManagementRouting every task to the right-sized model tier, opening safe tool-building to analysts, and handing issue triage to an autonomous agent.
Overview
An AI-native wealth-management firm was scaling its internal engineering and project-management operations faster than it could govern them. Canvas Labs Partners built an internal operations layer that routes model calls by task size, gives non-engineers a sandboxed environment for building their own tools, and runs issue triage and acceptance-criteria drafting as an autonomous agent.
40%+ lower LLM spend
model routing, enforced prompt caching and repo-wide token tracking
Analysts ship their own tools
sandboxed containers under automated CI quality gates
Issue triage runs unattended
an agent sizes effort and compiles acceptance criteria
Challenge
Every internal AI agent the firm shipped made LLM spend harder to predict, and there was no repo-wide view of where the money was going. Financial analysts had no safe route to building the internal tools they needed themselves, and project managers were burning hours on triaging issue channels and writing acceptance criteria by hand.
Solution
Canvas Labs Partners built a model-routing layer that sends each task to the right-sized model tier, so routine work no longer runs on the heaviest option available. Prompt caching is enforced everywhere rather than applied case by case, and a repo-wide token-tracking framework measures consumption across the codebase, catching leakage before it reaches the bill.
The second layer opens tool-building to non-engineers. Financial analysts work inside a sandboxed, containerised AI-assisted development environment stocked with custom LLM context and skill libraries. Docker holds each workspace apart, and automated CI quality gates govern what leaves it, so analysts build and maintain their own internal tools safely rather than the work landing on the engineering team.
The third layer is an autonomous project-management agent wired into Slack and Asana. It parses the issue channels where problems are first reported, sizes the effort involved, and compiles detailed acceptance criteria with screenshots attached, so a ticket arrives in Asana already specified rather than waiting on a project manager to write it up by hand.
Results & benefits
LLM infrastructure spend fell by more than 40 percent once routing, caching and token tracking were in place, and spend that had been unpredictable became something the firm could account for. Between 20 and 35 percent of engineering bandwidth came back, because the internal tools analysts needed no longer had to be built for them by engineers. Administrative overhead on the project-management side dropped by 80 percent, with issue triage, effort sizing and acceptance criteria compiled by the agent instead of written by hand, and tickets reaching Asana already specified.
Customer
AI-native wealth management firm
Services
- Agentic AI consulting & development
- LLM cost & model-routing engineering
- Internal developer tooling
Tech stack
About us
Canvas Labs Partners is a software consulting team that designs and ships production agentic AI systems: multi-agent pipelines, tool-calling architectures, and human-governed automation, together with the trading infrastructure, blockchain, mobile, and product engineering those agents plug into. We publish our work as a case ledger rather than a portfolio, so every claim on this site links to the engagement behind it. Engagements follow the same four-step discipline, Map, Bound, Instrument and Govern, whatever the stack, and every entry in the ledger carries a result we are willing to be measured against.
Delivered individually, in a prior role, by one of Canvas Labs Partners’ founding team members. Not by the firm itself.
Have a system like this one?
We built this one end to end. If you are facing something similar in wealth management, book a free discovery call and we will tell you what it would take.