Agentic AI is changing the way engineers work.
Take the Advisor360 engineering team, for example, which works in what Venkat Peri, head of agentic AI, describes as an “agentic development lifecycle.”
“Typing is cheap now, so humans move up the stack to what actually matters — architecture, planning, and defining the acceptance tests and harness that encode ‘done,’ Peri said. “Agents write the code against those; when the tests go green, we're confident to ship.”
The team at fintech company Advisor360 isn’t alone in using AI agents to fuel their workflow. Seventy-nine percent of respondents in a PwC survey said that AI agents are being adopted into their companies’ workflows. Of those that have started using AI agents, two-thirds report increased productivity.
Built In spoke with Peri about how the fintech company uses AI to ship faster while maintaining quality — and the impact that has on the business.
Advisor360 is a fintech company that helps wealth management firms operate, scale and serve their clients.
What’s your rule for releasing fast — and what KPI proves it works?
The rule is simple — nothing merges unless it passes the harness. Our automated rig that drives the whole assembled system end-to-end under controlled conditions and scores the outcome, so a plausible-looking change that's actually wrong gets caught before a human ever looks at it. This is the heart of the agentic development lifecycle we work in, typing is cheap now, so humans move up the stack to what actually matters — architecture, planning, and defining the acceptance tests and harness that encode "done." Agents write the code against those; when the tests go green, we're confident to ship. Speed and safety stop being a trade-off because the safety is automated. The gate isn't just unit tests; it's live evals and a live end-to-end run promoted against hermetic synthetic data, so we're scoring real behavior, not just green checkmarks. The KPI I trust is the eval-gate pass rate on the merge pipeline — green gate plus coverage holds means same-day ship — and we've pushed 300-plus pull requests across the stack since June with that gate mandatory on every repo. The next frontier: we're expanding the agentic development lifecycle beyond engineering gates to measure outcomes, not just whether the code is correct.
Which standard or metric defines “quality” in your stack?
Two things, in a deliberate order. The floor is coverage — a diff-coverage bar of roughly 90 percent enforced in CI on every repo. But coverage is just the floor; the real standard is test depth — would this test actually fail if the behavior broke? We built a "test-quality" discipline into the stack that audits tests for exactly that: does it assert intended behavior, does it cover the failure modes, or does it just mirror the code. A green suite that can't fail isn't quality — it's theater.
Name one recent AI/automation shipped and its impact on the team or business.
We shipped an AI reviewer that runs on every pull request across the whole stack — plus a Slack bot the team can invoke on any pull request to get a full cross-repo review posted back in seconds. It doesn't replace human review; it catches the boring, expensive stuff first — contract drift, weak tests, missed edge cases — so people spend their attention on judgment, not chores. The impact: review bandwidth stopped being the bottleneck even as PR volume climbed.
