Workshops: Agentic Coding
How to Review AI-Generated Code Securely: Workshop
When agents write much of the code, review becomes the decisive quality and security filter. This workshop trains developers and security champions to review AI-generated pull requests efficiently and thoroughly, and shows how AI review tools and automated checks can sensibly lighten the human load.
Who is this workshop for?
- Senior developers with review responsibility
- Security champions and AppSec teams
- Tech leads and maintainers
- Teams with a high share of agent-generated pull requests
- QA and release owners
What you will take away
- Know the typical bug and vulnerability patterns in AI-generated code.
- Spot hallucinated or substituted dependencies and protect against slopsquatting.
- Structure reviews of large agent diffs so essentials are not missed.
- Use AI review tools and SAST to catch routine issues so humans can focus on logic and architecture.
- Agree on review rules and a definition of done for agent-written code.
Agenda
Module 1
What is different about AI code
- Typical patterns: plausible-looking logic errors, skipped edge cases, weakened tests
- Security flaws: missing authorisation, injection, insecure defaults, secrets
- Hallucinated packages and supply chain risks
Module 2
Structuring reviews
- Exercise: reviewing real agent pull requests
- Order: intent, tests, interfaces, details
- Getting the agent to produce smaller pull requests
Module 3
AI helping with review
- GitHub Copilot code review, Claude Code and Codex as reviewers
- Custom review prompts and checklists as project rules
- Limits of automated review
Module 4
Automated guardrails
- SAST, secret scanning and dependency checks in the pipeline
- Licence checks and SBOM
- Branch protection and mandatory reviews
Module 5
Process and accountability
- Definition of done for agent code
- Accountability: whoever merges owns it
- Metrics for review quality
Review is the new bottleneck
Agents produce code faster than people can read it. Teams often respond with superficial approvals because the code looks tidy and the tests pass. That is exactly where the most expensive mistakes arise.
This workshop focuses on something general agentic coding training rarely covers in depth: disciplined, efficient review of code you did not write yourself.
Tools we use
We combine human review with automated and AI-assisted checks.
- AI reviewers in GitHub, GitLab or your platform
- Static analysis such as Semgrep, CodeQL or SonarQube
- Secret scanning and dependency audits
- Project rules for Claude Code, Cursor or Copilot that enforce security guidelines upfront
Background
jocapps, the company behind neooutput, is ISO 27001 and ISO 9001 certified. Our AI Code Engineers review agent-generated code daily in client projects and share their checklists.
Frequently asked questions
How is this different from general secure coding training?
We focus on the specific patterns of AI-generated code and on the review process itself, meaning what happens between the agent and the merge.
Can AI reviewers replace human reviews?
They are a very useful first stage but do not replace judgement on domain logic, architecture and accountability. We show what a sensible division of labour looks like.
Do we work with real pull requests?
Yes. On request we use anonymised pull requests from your repositories or prepared examples with typical flaws.
Workshop
Get in touch
Shall we get your review process ready for agent-written code?
Tell us a little about your team – we will get back to you within 24 hours with a proposal for content, format and dates.
Prefer to talk?
More workshops
- Workshops: Agentic CodingClaude Code for Non-Developers: Workshop for Product, Design & Ops
- Workshops: Agentic CodingTrain the Trainer: Scaling Agentic Coding Across Your Company
- Workshops: Agentic CodingVibe Coding Workshop for Founders & Business Teams
- Deploy, Convert & ScaleSEO for AI-Built React SPAs: Prerendering vs. SSR
