Workshop Slides
Navigate the deck covering models, agents, Codex workflows, and collaboration patterns.
Welcome & Goals
Set expectations for pairing with AI copilots in production codebases.
- Workshop promise: deliver safe, review-ready diffs with Codex/GPT-5.
- Success metrics: confident prompting, guardrails, and collaboration habits.
- Quick poll: Which LLMs and CLI agents do you already use on the job?
Objective Setting
Audience Poll
Agenda Preview
LLM Model Landscape
Compare today’s flagship models through a production lens.
GPT-5
- Strengths: broad tool plug-ins, high reasoning depth, enterprise governance.
- Watch-outs: cost tiers, rate limits, context strategy for large repos.
Claude
- Strengths: long context windows, alignment guardrails, friendly diff output.
- Watch-outs: streaming availability, region constraints, tool ecosystem.
Gemini
- Strengths: multimodal analysis, tight GCP integration, lower latency tiers.
- Watch-outs: auth scopes, billing per project, variance across runtimes.
CLI Agent Roundup
Pick the right command-line interface for your team’s workflow.
- Codex CLI: deep repo context, apply_patch, git-aware plans.
- Claude Code: realtime conversational loop, great for refactors.
- Gemini CLI: GCP-native auth, multimodal inputs, notebooks.
- OpenCode: lightweight OSS agent, customizable tool chain.
Compare permissions, audit trails, and extensibility. Encourage teams to standardize on two primary agents to reduce cognitive load.
Codex Deep Dive
Demonstrate the end-to-end Codex pairing loop.
- Plan stage: structured prompts, acceptance criteria, risk notes.
- Implementation:
plan → prompt → apply_patch → verify
workflow. - Review: inspect diff, capture rationale, prep PR checklist.
- Rollback & iterate: stash partial drafts, branch hygiene practices.
Live Demo
Prompt Templates
Guardrails
Environment & Tooling
Ensure every participant has a safe, productive setup.
- Repo baseline: test data, feature flags, seed scripts.
- Secrets management:
.env
templates, rotated tokens, audit logging. - Developer ergonomics: shell aliases, editor integrations, tracing dashboards.
- Observability: capture Codex session logs for retro and compliance.
Context, Sessions & Compaction
Keep Codex effective as repositories grow.
- Session hygiene: reset between tasks, document prompt history.
- Context tactics: targeted file excerpts, doc summarization, retrieval helpers.
- Compaction: trim resolved threads, archive logs, share mini-briefs.
- Troubleshooting: identify stale state vs. hallucinations quickly.
Working in Parallel
Coordinate multiple engineers collaborating with Codex simultaneously.
- Session ownership rules and branch etiquette.
- Playbooks for dividing a feature into AI-assisted subtasks.
- Conflict resolution: diff comparison, manual merges, shared prompts.
- Communication loop: daily sync snippets + Codex log highlights.
Working Asynchronously
Extend Codex sessions across time zones and schedules.
- Capture transcripts and artifact bundles for hand-offs.
- Automate nightly maintenance tasks with scheduled Codex jobs.
- Define review expectations when humans re-enter the loop later.
- Checklist for morning catch-up: diff digest, failing tests, next prompt.
Wrap-Up & Commitments
Close with clear actions and resources.
- Summarize the model/agent decision matrix and Codex workflow.
- Share resource links, office hours, and internal enablement channels.
- Ask each attendee to document one experiment to run in the next sprint.
- Collect feedback while the experience is fresh.
Slide 1 of 9