Same five engineers, ~19× the weekly output
From ~7 to 135+ sustained, with a best week of 160. Measured in Tovuti’s own sprints.
Even a junior engineer went from 2–3 approved enhancements a sprint to nearly all of them.
Monday through Thursday by late summer 2026.
- Engineering team
- 5 engineers
- Engagement
- Blueprint Equity Build Team, Q2–Q3 2026
- Focus
- Agentic development lifecycle, QA, codebase context
Tovuti didn’t need more engineers. It needed a different way to build.
Tovuti was rebuilding a mature learning platform on top of a legacy codebase, with a small team and a roadmap customers were waiting on. Historically, most engineering time went to fixing bugs, and a single bug could take weeks to trace. New mobile, learner and reporting experiences kept slipping, because every sprint surfaced problems no one knew were buried in the old code.
The team had started using AI to write code faster. The bigger opportunity was further upstream: letting agents do the work of understanding the legacy system, planning against it and checking the result, so engineers could spend their time directing the work.
The old belief
AI is a faster autocomplete.
Give each engineer a coding assistant and expect a modest bump in output.
The reframe
AI is the development lifecycle.
Agents audit, plan, build and verify in parallel. Engineers set direction, review and ship.
The gains don’t come from writing code faster. They come from removing the unknowns before a sprint starts.
Where AI went into the workflow
Tovuti’s engineers built the harness. Blueprint’s Build Team identified where AI belonged in the workflow, pair-programmed with the team, and added the context, QA and guardrail layers that let it run at full speed.
Audit the legacy code
A multi-agent audit crawled the legacy app, including tenant, admin and feature-flag behavior, and wrote parity documents for the rebuild.
“Map the monolith before you touch it.”Plan from parity docs
A planning step compares parity docs, the new code, the API and the designs, then writes tickets and design briefs for whatever is missing.
“Find the gaps before the sprint does.”Build in parallel
Engineers claim isolated workspaces and run multiple agents at once. Durable task specs let work survive model switches and context resets.
“One engineer, several agents, no collisions.”Verify before shipping
The new build is compared against the legacy site, and model-to-model code checks plus automated security passes run before anything ships.
“Let a second model check the first.”Give agents context
Agent instruction files keep specs and architecture decisions current, and a codebase knowledge layer serves agents only the context each task needs.
“Agents are only as good as what they can find.”Set the guardrails
A code-only policy for AI tools, customer data stripped from tickets before it reaches a model, and added capacity for heavy agent runs.
“Code goes in. Customer data doesn’t.”When it moved
The step-change came in July and August, as the team adopted the harness and the context, QA and guardrail layers went in behind it.
Source: Tovuti, as reported in Blueprint Equity check-ins and the Q3 2026 portfolio review.
Show the numbers
| When | What happened |
|---|---|
| Jul 2 | Major feature enhancements per sprint rose from two to five or six as more engineers worked with AI agents. |
| Late Jul | Inbound engineering tickets fell to a record low for a single week. |
| Aug 10 | With the audit, plan, build and verify harness running parallel agents, the best week reached 160 story points, about 23× the old run-rate. |
| Aug 27 | Production releases Monday through Thursday. The mobile app was submitted to the app stores and the reporting MCP server was complete. |
| Sep 17 | Model-to-model code checks run before anything ships, and nearly every new product surface was in QA or staging. |
| Oct 20 | The same five-person team sustaining 135+ story points a week, about 19× where it started. |
What changed
Within one quarter, the same five-person team was shipping at a different order of magnitude.
Source: Tovuti’s own sprint data, reported in August and October 2026. Story points are relative to the team.
Show the numbers
| Story points a week | |
|---|---|
| Before the harness, H1 2026 | 7 |
| Best week, August 2026 | 160 |
| Current run-rate, October 2026 | 135 |
- The roadmap cleared. Mobile, a new learner experience, assignments, reporting and a standalone AI course builder all reached QA or release in about ten weeks, after slipping for a quarter.
- QA stopped being the bottleneck. Model-to-model code checks caught bad code before it shipped, without adding QA headcount.
- Inbound engineering tickets fell to a record low.
- New AI products followed. An MCP server for reporting shipped, and the team moved straight on to its next generation of AI features.
Before
~7 story points a week.
Most engineering time spent on bug fixes. Hidden legacy behavior surfacing mid-sprint. A roadmap slipping by quarters.
After
135+ story points a week.
Unknowns mapped before the sprint starts. Code checked before it ships. Releases four days a week on the same team.
Our process is much more aligned and solid. We’re doing five to six major feature enhancements on every sprint.
Matt Baker, CEO, TovutiWhat it meant for the people doing the work
The engineer
The builder
Rational
- Legacy behavior documented before work starts.
- Several agents working in parallel.
- Code checked before review, not after release.
Emotional
- Early worry that AI would replace the role.
- Confidence once the tools proved useful.
- More building, less firefighting.
CEO and product leadership
The operator
Rational
- A roadmap that ships on dates customers can plan around.
- Output scaling without adding headcount.
- Capacity freed for new AI products.
Emotional
- Confidence putting new surfaces in front of customers.
- Out of the “we can’t build that yet” conversation.
- A team that sees what it can do.
What we saw across the team
Adoption wasn’t uniform, and that’s the pattern worth copying. The highest-output engineers run around ten agent sessions in parallel and keep experimenting. Solid engineers who stop at two or three sessions get a fraction of the gain. Closing that gap is mostly about enablement and permission, not tooling.
The lesson. Don’t stop at giving engineers an AI assistant. Put agents into the lifecycle (audit, plan, build, verify), give them context and guardrails, and your existing team can ship an order of magnitude more.
How to read the numbers. Story points are Tovuti’s own sprint measure and are relative to the team, so they are paired here with the products that actually shipped.