We gave PortableMind agents one job: improve our own production platform every night. They audited it, fixed it, tested it and handed us a merge request to approve over breakfast. Here's what 100 nights of that did.
Every number on this page comes from our own production records, June 20 – September 27, 2026.
One night
One orchestration and a team of specialist agents. Each has a defined job, and none of their changes lands without a person approving it.
An auditor agent scans for security advisories, outdated packages, duplication and design problems.
It writes a short backlog (4 tasks max) into a tracked project, with exact files, risks and acceptance checks.
Developer agents work in parallel, each on its own branch with its own merge request.
A tester agent runs the full suite (20,000+ tests). A regression sends the work back.
One merge request and one test report are waiting for your review. Nothing merges without you.
Before and after
The first audit scored our backend 4/10 and listed critical findings. The latest one found no security advisories, only small, precise cleanups like a 15-line helper copied into three files.
Test examples per tester-agent report, June 23 – September 27, 2026. Feature work shipped on the same branch over the same period.
Why it compounds
A scheduled script runs the same check every night. PortableMind runs an orchestration that picks up from where the previous nights left off.
Specialist agents hand work to each other through defined stages, with bounded scope, budget limits and a human approval gate at the end.
Every one of the 391 tasks lives in one project, and every run saves its audit, merge request and test report. You can trace any change back to the reason it was made.
Each night's audit reads the full task history, avoids work that's still awaiting review, and carries tonight's leftovers into tomorrow's plan.
“Collision check: the previous run's integration branch is not yet merged … Tonight's tasks avoid all of those files and that bump. … Candidates cross-checked against the prior ~380 backlog tasks to avoid repeats.”
The auditor agent, in its own report, September 27, 2026
Beyond code
Code was our first case study because we could measure it exactly. The same pattern (audit, plan, act, verify, approve, remember) works for any recurring work that never makes it onto the calendar.
Re-check documents against the latest rules every week.
Find stale pages, draft updates and route them for approval.
Reconcile records, flag anomalies and open tracked tasks.
Triage, draft replies and escalate with full context.
Second-time founders. Built for ourselves first.
Rick has spent 35+ years leading engineering teams and implementing enterprise architecture at scale. He co-founded TrueNorth with Russell, and today he keeps PortableMind's architecture grounded in enterprise best practice while holding the line on a simple, concise UX: the surfaces teams already know, wired to one source of truth.
Rick on LinkedIn
Russell co-founded TrueNorth and led its engineering, building the rules and workflow engine that carried the company to its 2019 acquisition by one of the largest healthcare payers in the country. He stayed on as VP of Engineering to run implementation and integration. An open-source contributor, he now architects the PortableMind platform: the APIs, shared memory, tenancy and agent orchestration under everything you see.
Russell on LinkedInBuild your first orchestration on PortableMind. Agents do the work overnight, every step is tracked, and nothing ships until you approve it.
Start free on PortableMindWant the full story? Read the white paper: “The Codebase That Gets Better Every Night.”
