AI coding agent adoption for professional software teams

Train your developers to use Claude Code and Codex without turning delivery into guesswork.

Mostly Harmless Code helps engineering organizations turn AI coding tools from individual experiments into repeatable team workflows: scoped tasks, test-backed changes, review discipline, and practical security boundaries.

Book intro call View programs

1 day: hands-on workshop format
2-4 weeks: team enablement sprint
2 tools: Claude Code and Codex

$ codex "triage flaky auth tests"

reading repo context...

found failing path: session refresh edge case

$ claude "prepare minimal patch with tests"

patch ready: 3 files, 2 tests added

review: security boundary unchanged

The enterprise problem

AI adoption is already happening. The question is whether the team learns it deliberately.

Individual experimentation

private prompts and inconsistent tool use
large unreviewable diffs
unclear privacy and source-code boundaries
managers cannot tell what is working

Team operating model

shared workflow patterns and task sizing
test-backed changes and reviewable patches
explicit rules for secrets, data, and policy
measurable adoption with practical guardrails

Training products

Offer formats that match how large engineering organizations buy training.

Executive Briefing

For CTOs, VPs, directors, and enablement leaders deciding how to roll out AI coding tools.

Duration: 90 minutes
Output: adoption map, risks, next-step options

Developer Workshop

Hands-on training for professional developers using Claude Code and Codex on realistic repo tasks.

Duration: half day or full day
Output: repeatable workflows, prompt patterns, review checklist

Team Enablement Sprint

A deeper engagement that helps teams apply the workflow to their own codebase and delivery process.

Duration: 2-4 weeks
Output: playbook, internal champions, adoption metrics

Curriculum map

Teach developers the loop, not a pile of prompt tricks.

Foundations

Tool model, context windows, repo navigation, task decomposition, agent limits.

Workflows

Bug triage, test repair, refactoring, code review preparation, documentation updates.

Controls

Secrets, IP, dependencies, generated code review, audit habits, team policy.

Measurement

Cycle time, review quality, test evidence, rework, developer confidence.

Claude Code

Codex

Deep pairing sessions, broad refactors, interactive repo work, local planning.

Shared engineering discipline

Task execution, automation, codebase navigation, tests, review-ready changes.

Enterprise readiness

The training has to satisfy engineers, managers, security, and procurement.

Security posture

Training covers secrets handling, repo boundaries, customer data, tool permissions, and review obligations.

Manager visibility

Leaders get a practical adoption model: what to measure, which teams to pilot, and how to avoid cargo-cult usage.

Procurement path

Clear formats, delivery modes, prerequisites, cancellation terms, and NDA-friendly training options.

Real engineering work

Exercises focus on patches, tests, reviews, and system understanding rather than generic chatbot demos.

Rollout roadmap

A concrete path from curiosity to managed adoption.

1

Assess

Map teams, current tool use, repo constraints, and policy concerns.
2

Pilot

Run workshop with one or two teams and gather workflow evidence.
3

Standardize

Publish team playbook, review expectations, and safe usage rules.
4

Scale

Train champions, expand by team type, and measure delivery impact.

Prototype stack decision

Static first, framework later if the content starts behaving like an app.

Now: static HTML

Best for fast iteration on message, diagrams, offer structure, and executive polish.

portable to Vercel or Cloudflare Pages
no build chain to debug
easy to hand-edit while the offer is still changing

Next: Astro

Best when articles, reusable layouts, RSS, tags, and SEO metadata become regular work.

static output with content structure
clean migration from this prototype
less app overhead than Next.js

Later: Next.js

Best if the website grows into a product surface with forms, dashboards, auth, or dynamic data.

component system and app behavior
more moving parts
use when there is real product complexity

Next step

Book an intro call or request the syllabus.

The launch version will use Google Workspace for business mail at hello@mostlyharmlesscode.com and a dedicated booking link once the scheduling account is created.

Email hello@ Request syllabus