Sessions

Our monorepo CI took 40 minutes on a good day. This talk walks through how we cut it to 6 minutes with content-addressed caching, remote execution, and a test-selection model — including the two migrations that failed first. You'll leave with a decision framework for which incremental-build investments pay off at which repo sizes, and the graphs to convince your platform team. Updated: now includes 2026 benchmark data. This session now includes a live demo of remote build caching.

Treating every existing API surface as an agent tool: the adapter patterns, the auth problems, and what breaks at scale when agents call APIs designed for humans. Includes a rubric for deciding which endpoints to expose and which to wrap.

Ten years of developer workflow change compressed into forty-five minutes: what actually moved the needle, what was fashion, and where the next decade of leverage is hiding. A data-heavy look at build systems, editors, CI, and the arrival of agents that write and review code alongside us.

Coverage numbers say a line ran, not that a feature works. A practical tour of functionality-level browser testing and what it catches that unit tests never will, with a migration plan for suites that are already too slow.

Our monorepo CI took 40 minutes on a good day. This talk walks through how we cut it to 6 minutes with content-addressed caching, remote execution, and a test-selection model — including the two migrations that failed first. You'll leave with a decision framework for which incremental-build investments pay off at which repo sizes, and the graphs to convince your platform team.

What team structure, rituals, and career ladders look like when a meaningful share of the work is done by agents rather than people. Two reorgs, one of which was a mistake, described honestly.

Median build time is the number people quote; p99 is the number that ruins afternoons. How we found the tail, what caused it (spoiler: not the compiler), and the four changes that halved it.

A hands-on two-hour workshop building a minimal remote build execution service: content-addressed storage, action caching, and a worker pool. Bring a laptop with Docker installed.

Internal platforms fail politely: nobody complains, everybody routes around them. A field guide to platform adoption metrics, paved-road design, and the deprecations that make a platform credible.

Measured throughput changes across a large engineering org after AI tooling rollout — what moved, what did not, and which measurements were misleading enough that we stopped reporting them.

A frank panel on enterprise agent deployments: governance, budgeting, evaluation, and the organisational antibodies that kill promising pilots. Audience questions encouraged and unfiltered.

If your platform team has no churn metric, it has no users — it has hostages. A keynote on running internal infrastructure with product discipline, including the uncomfortable parts about pricing and sunsetting.

How a 12,000-line hand-rolled subsystem was replaced by a 200-line agent skill, what broke along the way, and the decision framework for when this trade is worth making.

Cost structure, procurement realities, and measured returns of trillion-token RL deployments inside large regulated enterprises. What the finance team asks that the research team never anticipated.

Dual writes, shadow reads, backfills that take a fortnight, and the rollback you will need at 2am. A migration playbook derived from four production cutovers, two of which went badly.

A post-mortem across enterprise agent deployments: the budgeting, governance, and ROI-modelling mistakes that sink them before the technology ever gets a chance to fail on its own merits.

A two-hour workshop on building a developer productivity dashboard your engineers will trust: which signals to collect, which to refuse, and how to publish the methodology alongside the numbers.

Everything about a monorepo that is fine at 200 engineers and unbearable at 4,000: code ownership, dependency graphs, review routing, and the tooling investments that bought us another order of magnitude.

Documentation is consumed by agents now, and they are unforgiving readers. Structuring docs so both humans and machines can navigate them, with measurements from a year of doing it.