TEXO
Overview
When agent cycle time spiked 15x, the dashboard said failure. The work systems said otherwise.
- Systems integrated
GitHub
Jira- Agents instrumented
Claude
Codex
Copilot
Cursor- Team
- less than 1000 developers
- Stakeholder
- Chief Technology Officer
- Process
- AI Software Development Lifecycle
Highlights
- Adoption tracked by tool and by model: no agent sessions in May, rising to 67% of the platform team by July
- Every agent session tied to the work it touched, down to AI code share on each pull request, ranging from 8% to 96%
- Human-only cycle time fell 49% over four months. Agent-assisted task volume rose 5×.
- On support work, agent-assisted cycle time fell 68% and converged with human-only work, the signal that the agent is handling comparable complexity
Situation
TEXO is a private equity-backed software vendor for the commercial laundry and linen-services industry, growing through both organic expansion and acquisition. It had started putting agents into its engineering workflows, with Claude, Codex, Copilot, and Cursor in use across its platform engineering and support and maintenance workstreams. Adoption was early and uneven, and the CTO wanted it to grow.
Challenge
Growing adoption meant knowing where it stood: which developers used agents, with which tools, and on what work. It also meant knowing what agent-assisted work looked like once it reached GitHub, how much of the merged code agents wrote, and whether that work finished faster than the same team's work without an agent.
The agent tools showed usage. Not who used them, on what work, or whether it finished faster than the same team without an agent. A metric moving the wrong way with no way to explain it is how AI programs get cancelled.
Solution
Bloomfilter connected to the Jira and GitHub instances TEXO already ran and mined the event streams they already produced. No new workflows, no manual mapping. End-to-end flow became visible, including where work moved backward and what it cost.
Bloomfilter then overlaid agent activity from all four tools on that same structure. A developer binds a session to a Jira ticket, and the session inherits the pull request, build, and deployment that ticket already correlates to. Capture runs through an open-source plugin, with prompt contents excluded and nothing leaving the boundary TEXO had already approved.
With sessions tied to work, three comparisons became possible: adoption by tool and model, AI code share on every pull request, and cycle time for human-only work measured against human-plus-agent work.
Outcome
Adoption became a number the CTO tracks
No agent sessions in May. Adoption climbed through June and reached 67% of the platform team by July. Average turns per session more than doubled, from 2.7 to 6.5, and held there. Engineers were working with the agents, not firing one-shot prompts. File-edit volume kept climbing through July, the month the team stopped experimenting.
| Month | Adoption rate | Avg turns / session | Session growth (vs. prior month) |
|---|---|---|---|
| May | 0% | — | — |
| June | 50% | 2.69 | First sessions |
| July | 67% | 6.32 | +183% |
| August | 50% | 6.55 | −46% |
The metric that looked like a problem
In August, agent-assisted cycle time jumped from 1.3 days to 21.0 days. A 15x increase, in one month, on the flagship AI initiative. On a usage dashboard, that is where the program ends.
With sessions tied to work, TEXO could see the other half of the picture. In the same month, human-only cycle time dropped to its lowest point in four months, and the team completed five times more agent-assisted tasks. The agents had not slowed down. They had been tasked with more complex work that takes more time.
| Month | Agent-assisted share of completed work | Agent cycle time | Human-only cycle time |
|---|---|---|---|
| May | 0% | — | 9.75 days |
| June | 11% | 6.8 days | 7.98 days |
| July | 22% | 1.33 days | 6.98 days |
| August | 54% | 21.0 days | 4.97 days |
In June, agents fixed small things. By August, they were building platforms.
The team did not run more sessions. It put agents on harder problems. Average story points on agent-assisted work climbed month over month, and the nature of the work changed with it.
By August, agents were co-authoring security infrastructure and multi-week architectural features. The cycle time spike was not a measure of inefficiency. It was a measure of ambition.
| Month | Representative agent work | Avg points | Nature of work |
|---|---|---|---|
| June | NuGet package updates, AI rules setup, automated testing | 1–3 pts | Maintenance and tooling |
| July | AMS proxy migration (C#/.NET), CI/CD refactor, testing migration | 1–8 pts | Refactoring and infrastructure |
| August | Workflow execution engine, WAF security, agent trust hardening | 3–8 pts | Core platform architecture |
When an agent writes 96% of a pull request, the engineer becomes the architect
AI code share ranged from 8% to 96% depending on the task, and the highest shares landed on the most complex greenfield work. On those pull requests the agent was the primary author and the engineer was the architect, the reviewer, and the decision-maker. Bloomfilter measures that split per pull request, so the ratio is a fact about the merge rather than a claim about the tool.
| Pull request | AI code share |
|---|---|
| Feature/all-1089 fixes | 96.1% |
| Stabilize execution pickers | 89.6% |
| Fix search for accounts and workflows | 87.4% |
| Feature/permission fixes | 69.3% |
| Refactor AccountService | 66.1% |
| Feature/agentic documentation | 62.2% |
| Feature/all-1089 implement missing steps | 32.6% |
| Raise AMS test coverage to 80%+ | 8.5% |
On support work, the same instrumentation showed a different shape
The support and maintenance workstream ran a steadier pattern across all four tools over four months. Agent-assisted cycle time fell 68% from its June peak. Throughput rose 53%. Velocity rose 22%.
By August, agent-assisted cycle time (11.1 days) had converged with human-only cycle time (10.01 days). The agent was carrying work of comparable complexity to what the team handled without it. Convergence is what maturity looks like, and it is only visible when both numbers come from the same delivery record.
| Month | Agent cycle time | Human-only cycle time | Task volume (vs. May) | Velocity |
|---|---|---|---|---|
| May | 13.9 days | 9.95 days | baseline | 50 pts |
| June | 34.7 days | 19.85 days | +19% | 49 pts |
| July | 18.3 days | 12.1 days | +36% | 61 pts |
| August | 11.1 days | 10.01 days | +53% | 61 pts |
What the instrumentation changed
Seeing the spike meant TEXO could hold course through it. Seeing what the agents were working on meant the CTO could steer adoption toward the hardest problems instead of mandating it across the board. The lesson was not that agents make everything faster. It was that agents let the strongest engineers take on the problems they had been routing around.
49%
Reduction in human-only cycle time, May to August
68%
Reduction in agent-assisted cycle time on support work
About the data
Based on agent session data captured via Bloomfilter, May 1 – August 31, 2026, across TEXO's platform engineering and support and maintenance workstreams. All figures reflect completed work items and attributed agent sessions only. Unattributed sessions are excluded, so actual AI usage was likely higher. Cycle times are medians. Adoption rate is measured against the team using agents, not the full engineering organization.