Day 1 · Recorded 6 May 2026
Background agents for genomics, cloud operations, and self-evolving skills
Genomics platforms combine scientific domain knowledge, domain-specific file formats, and cloud jobs that can fan out across thousands of instances. Xiucheng Quek shows how Genentech uses secure background agents to triage failed analyses, generate operational reports, automate routine development, and uncover why highly specific skills can make agents slower.
Xiucheng Quek, Principal Scientific Software Engineer, Genentech, Inc
What's in this session
Genomics is not a normal data-pipeline problem. A platform team has to operate across bioinformatics tools, scientific workflows, cloud infrastructure, and failures that require both domain and engineering judgment.
In this session, Xiucheng Quek walks through how a small Genentech team started with safe operational work, then expanded into failed-analysis triage, incident reports, daily summaries, routine development tickets, direct merges into controlled environments, and nightly self-evolution through logs, Slack updates, and Confluence notes.
For engineering leaders, the lesson is environment first, autonomy second. Secure infrastructure, tests, canaries, and clear operating boundaries make background agents useful; the Specificity Paradox shows why generic skills that teach agents what to inspect can outperform rigid checklists that prescribe exactly what to find.
Inside the recording
- 00:00 Background agents for genomics platforms
Quek introduces agents operating across bioinformatics, cloud infrastructure, and research workflows. - 03:00 Choose the right environment, then start small
The talk's core pattern is secure infrastructure first, then incremental autonomy. - 06:00 When cloud-scale genomics jobs fail
Agents help triage analyses that can span hundreds or thousands of instances. - 09:00 Incident reports and operational agents
With the right instructions, agents can turn triage work into high-quality reports. - 12:00 Letting the bot merge into main
Tests, canaries, and controlled environments make hundreds of routine tickets safe to automate. - 15:00 Self-evolving skills and the specificity paradox
Generic skills often outperform detailed checklists because they keep diagnosis flexible.
More sessions on agent infrastructure
- Building Minions: agents on a 30-million-line codebase — Alistair Gray, Stripe
- Building a company-internal background agent system — Cole Murray, Open Inspect
- From Assisted to Delegated: Cloudflare's AI Engineering Stack — Rajesh Bhatia, Cloudflare