How Product & Technology used AI across everything we shipped in the last 14 days — cross-referencing Jira (source of truth for what we worked on) against Claude Code telemetry, GitHub, Confluence and Slack.
Jira is the source of truth (last 14 days, 16-30 June). Claude Code tokens are scoped to company repositories: 5.92B of 17.12B total Claude Code tokens (35%) were on company repos - the rest is planning, research, POCs (marketplace/Cowork) and analysis. Spend is never shown.
Jira is the source of truth. Below: where the work landed and the mix of work types across all active projects in the window. The hatched (diagonal-slash) part of each project bar marks the issues that involved AI (estimated from the project's AI-coding signal).
| Project | Focus | Issues | AI coding signal |
|---|---|---|---|
| QAFNDN | QA: Foundation (FNDN) | 467 | Partial |
| FIT | Foundation & Integration | 134 | Strong |
| FARM | Farm And Ranch Management | 132 | Strong |
| CS | Customer Support | 78 | Partial |
| DP | Data Platform | 76 | Strong |
| PS | Production Support | 37 | Partial |
| INFRA | Infrastructure - DevOps & Infra | 36 | Strong |
| RND | Research & Development | 30 | Partial |
| PRD | Product Operations | 29 | Partial |
| QACEX | QA: Cattle Experience (CEX) | 13 | Partial |
| TEM | Tempo | 13 | None |
| AR | Access Requests | 13 | None |
| TAGS | Tagging Development | 9 | Partial |
| GRAS | Grassland & Environmental Mgmt | 2 | None |
| DDA | Data & Analytics | 1 | Strong |
AI is now applied at every stage of delivery. The colour bar shows how well each stage is currently supported — green is well-covered, amber is partial and an opportunity.
Measured Claude Code usage over the last 14 days (Honeycomb OpenTelemetry), scoped to sessions that edited files in a recognised Aiske.ai repository — i.e. real Jira/GitHub project work. Planning, research, agentic exploration, Cowork POCs and analysis are excluded. All figures are for the window only — not cumulative over the full codebase or all time.
| Team | Focus | Tokens (company repos) | Lines of code | What AI is used for |
|---|---|---|---|---|
| Architecture & DevSecOps | Backend services, MuleSoft/iPaaS integrations & the AI code-review pipeline | 1.86B | 35,631 | Backend services, MuleSoft/iPaaS integrations & the AI code-review pipeline (1 active dev) |
| Foundations & DevSecOps | Mission Control, infrastructure/DevOps, platform & shared services | 1.73B | 48,245 | Mission Control, infrastructure/DevOps, platform & shared services (5 active devs) |
| Data & Intelligence | Datalake / hwlake-bronze models, lineage catalogue & data pipelines | 1.34B | 29,315 | Datalake / hwlake-bronze models, lineage catalogue & data pipelines (2 active devs) |
| Core Solutions | FARM & Aiske features, mobile app, tests and the AI part of the app | 0.96B | 24,139 | FARM & Aiske features, mobile app, tests and the AI part of the app (4 active devs) |
| Product Delivery (direct) | Cross-team product delivery & tooling | 0.03B | 0 | Cross-team product delivery & tooling (1 active dev) |
| Network & Integrations | Integration engineering & connectors | 0.00B | 82 | Integration engineering & connectors (1 active dev) |
| Epic | Project | How AI was used | Status note |
|---|---|---|---|
| FIT-8530 Regression 99.0.0 | FIT | code + specs | To Do |
| DP-964 Star ranking, Dashboards, Superset clean up | DP | code + specs | To Do |
| RND-34 Integration Unification: unified ACL + strangler rollout | RND | code + specs | In Progress |
| FARM-11863 QA / Farm Assurance Tracker — Automated Eligibility Tracking (All Markets) | FARM | code + specs | In Progress |
| FARM-12136 Aiske 1.5 Weight Upload MVP | FARM | code + specs | To Do |
| FIT-8537 Sync Manager: improve logs, reports and reporting dashboard | FIT | code + specs | In Progress |
The most important section: concrete, evidence-based moves to push productivity further. Ordered by impact.
QA: Foundation (FNDN) was the single busiest project this fortnight - 467 issues, including 324 test executions and 166 test plans, almost all run and maintained by hand. This is the clearest AI win available: have Claude Code maintain Cypress/E2E specs in lockstep with UI copy and auto-draft test executions from each test plan. Testing is the weakest AI-assisted stage in our lifecycle strip.
FARM and FIT together carried 266 Jira issues, yet Core Solutions logged the lowest scoped Claude Code usage of any engineering team (0.96B tokens, 4 active devs) - and a chunk of the team's fortnight went into a marketplace/Cowork POC that sits outside our tracked repositories. Redirect that momentum onto FARM/FIT tickets in ngx-app and de-svc-herdchat, and pair the team's heaviest adopter with the lighter ones on real tickets.
Customer Support (78), Production Support (37) and Product Operations (29) generated real Jira volume this fortnight but show little to no AI-coding signal because the work is triage, reproduction and comms rather than commits. Give these teams AI templates for ticket triage, reproduction summaries and first-draft customer-facing notes so the assist shows up where the work actually is.
Only 14 developers were active in Claude Code on company repositories this fortnight, against a larger P&T engineering headcount. Two teams (Network & Integrations, Product Delivery direct) each had a single, very light active user. Use the leaderboard to name who sits at levels 1-3 and pair them with the four heaviest adopters on live tickets.
The iot-svc-ipaas repository accounted for ~21% of AI-written lines and a fifth of AI-created PRs this fortnight, but the Network & Integrations team reads as a single near-silent user - the integration work is being credited elsewhere. Confirm sign-in/repo attribution for that team so their genuine AI contribution is visible and they get proper coaching.
We saw ~60 AI-created PRs in the window (Mission Control and aiske-infra led), but squash-merges drop the trailer so GitHub attribution under-reads. A commit-hook that preserves the trailer makes AI contribution measurable without changing how anyone works - and lets us tie PR-level AI use back to Jira.
Backend and data services (iot-svc-ipaas, dp-svc-hwlake-bronze, de-svc-herdchat) produced the bulk of AI-written lines this fortnight, and several epics (DP-964 dashboards clean-up, RND-34 integration unification, FIT-8537 sync-manager reporting) are clearly phase-based. Scaffold phase 2/3 off phase 1 and reuse the patterns across DP/FIT/INFRA rather than re-deriving them each time.
With 37 epics active this fortnight, generate first-draft release comms and PRDs from each epic and its child issues - especially the customer-visible FARM work (FARM-11863 eligibility tracking, FARM-12136 Aiske 1.5 weight upload). It turns work we already tracked in Jira into shippable communication with almost no extra effort.
Sources: Jira (REST API, 16-30 June 2026, full paging), Honeycomb Claude Code OTel (company-scoped), GitHub / airbug AI review, AI cost-benefit model (Outcome estimate).
pt-ai-usage-report skill, a companion to the AI Adoption Leaderboard. Figures are decision-support; AI-coding telemetry covers opted-in Claude Code users. Irish/UK spelling throughout.