Workshops: Agentic Engineering Practices
Performance Optimization with AI Coding Agents: Workshop
Ask an agent to make code faster and it will happily rewrite it, often without making anything faster. This workshop teaches the opposite order: measure first, let the agent work on the real bottleneck, prove the gain with benchmarks and guard it against regressions. Your team practises on its own slow endpoints, queries and pages.
Who is this workshop for?
- Backend teams with slow endpoints, batch jobs or high cloud bills
- Frontend teams working on Core Web Vitals, bundle size and rendering
- Performance and reliability engineers
- Database-heavy teams with slow queries and ORM issues
- Tech leads who want performance to stay good once it is fixed
What you will take away
- Give the agent profiler output, flame graphs and query plans instead of vague complaints.
- Write reproducible benchmarks and load tests with the agent before touching production code.
- Judge agent-proposed optimizations critically: real gain, readability cost, correctness risk.
- Optimize database access, caching, algorithms and frontend loading paths with measurable results.
- Add performance budgets and benchmark comparisons to CI so regressions surface at review time.
Agenda
Module 1
Measure before you touch
- Why agents guess at bottlenecks and how to stop it
- Profilers per stack: flame graphs, sampling profilers, browser DevTools
- Exercise: profile a slow path and hand the evidence to the agent
Module 2
Benchmarks as the feedback loop
- Micro-benchmarks with tools such as JMH, BenchmarkDotNet, pytest-benchmark or Go benchmarks
- Load tests with k6, Gatling or Locust, written and maintained by the agent
- Stable measurement: warm-up, noise, comparing several runs
Module 3
Typical optimizations
- N+1 queries, missing indexes and execution plans
- Caching, batching and needless serialization
- Algorithms and memory allocations
Module 4
Frontend performance
- Bundle analysis, code splitting and lazy loading
- Rendering costs and Core Web Vitals with Lighthouse
- Images, fonts and third-party scripts
Module 5
Keeping it fast
- Performance budgets and benchmark comparisons in CI
- A review checklist for optimization pull requests
- Documenting decisions so the next refactoring does not undo them
Why “make it faster” is a bad prompt
Without measurements, coding agents optimize what looks inefficient, not what is slow. The result is clever, harder-to-read code with no noticeable effect, while the real cause sits in a database query or network call nobody looked at.
With evidence in context the picture changes. Given a profile, a benchmark and a clear target, Claude Code, OpenAI Codex, Cursor or GitHub Copilot can propose several candidate fixes, implement them on branches and run the benchmark for each. Your developers then choose with data instead of gut feeling.
Guardrails for optimization work
Performance changes carry their own risks: subtle concurrency bugs, stale caches, results that differ in edge cases. So the agent works under fixed rules:
- Existing tests stay green; correctness beats speed
- Every claimed gain comes with a before-and-after measurement
- No production profiles containing personal data in the agent's context (GDPR)
- Caching and concurrency changes always get human review
Bring your slowest code path
The day works best with two or three concrete performance problems from your systems: a slow report, an expensive query, a sluggish page. We prepare the measurement setup with you beforehand; where production data is involved, we use anonymised or synthetic datasets. Remote or on-site, in English or German.
Frequently asked questions
Which languages and stacks do you cover?
Java and Kotlin, .NET, Python, Node.js and TypeScript, Go, common web frontends and SQL databases. The principles are the same; we pick the tooling to match your stack.
Can the agent run the profiler itself?
Yes. In development and test environments agents can run profilers, benchmarks and load tests and analyse the results. For production we recommend read-only access through approved observability tools.
How do we know the speed-up is real?
Through repeated measurements under comparable conditions and a benchmark that stays in the repository. We show how to spot measurement noise and why a single run proves nothing.
Workshop
Get in touch
Which slow part of your system should be measurably faster after one day?
Tell us roughly how many people will join and pick dates that work – we will get back to you within 24 hours with a proposal for content, format and schedule.
Prefer to talk first?
15 minutes, free and without obligation – with our AI Trainers and AI Code Engineers.
Book a call for free (opens in new tab) +49 761 519 751 50 (24/7)Request a training / workshop
Workshop: Performance Optimization with AI Coding Agents: Workshop · 1 day
More workshops
- Workshops: Agentic Engineering PracticesAPI-First Design with AI Agents: OpenAPI & GraphQL Workshop
- Workshops: Agentic Engineering PracticesCoding Agents in Large Codebases & Monorepos: Workshop
- Workshops: Agentic Engineering PracticesGit Workflows for AI Coding Agents: Half-Day Workshop
- Workshops: Agentic Engineering PracticesSelf-Healing CI with AI Agents: Builds, Flaky Tests, Updates
- Workshops: Agentic Engineering PracticesTaking Over Vibe-Coded Prototypes: Workshop for Dev Teams
- Deploy, Convert & ScaleDeploying a Cursor-Built App to Production: A Practical Guide
