Kirill Chernakov's Talks

Make Claude unblock the 10x engineer inside you

· SIEM-eng all-hands · 22 min

Speed
Open slides Materials & links Download video

Additional materials

Talk: Make Claude unblock the 10x engineer inside you — SIEM-eng all-hands, 18 Jun 2026.

The skills repo

Install the shared Elastic skills once, then use them as slash commands in Claude Code:

gh skill install elastic/skills

Useful ones from the demo:

  • start-kibana — boot a local Kibana + ES instance from a PR or branch, with per-worktree isolated data and logs.
  • verify — drive the running app to confirm a change actually works, not just that tests pass.
  • dogfood — structured exploratory QA pass with reproduction evidence: screenshots, repro videos, and step-by-step notes.
  • pr-screenshots — capture before/after screenshots and update the PR description.

Source: elastic/skills. If you build something useful, PR it back.

The "ship it" cap challenge

Ship a sophisticated feature for under $10 in Claude tokens and I will send you a ship-it cap as a souvenir.

Ship-it cap challenge page

Submit your PR for the cap.

Rules:

  • Has to be a real merged PR; non-trivial bug fixes count.
  • Send me the PR link plus your /cost printout from the session.
  • Honour system on the token count.

The three takeaways, restated

  1. Ground the model, do not prompt-engineer it. Models are already smart. Your job is to give them hands, eyes, logs, and loops — tools to verify themselves. That is what unlocks autonomy.
  2. Fix the foundation first. Schemas, validation, error messages, docs. AI amplifies whatever foundation you have. Weak foundation means a million scripts and overengineering. Strong foundation means fast ship.
  3. Distill yourself. Have the agent read a year of your PR reviews and figure out what you care about: architecture, visuals, perf, whatever. Then it stops wasting your time on the parts you already check.

And the meta-takeaway: do not start by optimizing tokens. Burn them on purpose to distill yourself and accelerate the team. Optimize after, once you know what is worth keeping deterministic.

From the demo

  • Bug fix walked through mid-demo — workflow reformatting on enable/disable toggle: elastic/kibana#273930. Verification video and PR description generated by Claude end-to-end. Cost: about $23, most of which was waiting for Kibana + ES to start, not LLM latency.
  • Live agent run — issue from the teammate: elastic/security-team#17816. The PR the agent produced: elastic/kibana#274004.

Reach out

Happy to pair if you want to set this up on your own workflow.