A capable assistant and a hopeful prompt
The model can generate code, but the team still has to guess what it saw, what it was allowed to change and whether the result is safe.
Self-paced training for software engineers
Become the engineer who designs the delivery system.
A practical self-paced course for software engineers who want to design agentic delivery systems that can be trusted: explicit authority, current context, deterministic verification, model routing, controlled roles and evidence-gated delivery.
£150One-off payment per learner
9 lessonsPlus an optional implementation appendix
Return any timeSecure, self-paced portal access
What changes
Using an agent is easy. Engineering the system around it is the real work.
This course is grounded in runnable examples, deterministic gates and real operating decisions. You will move from ad hoc assistant use to a workflow with context, authority, evidence and ownership.
A governed agentic delivery loop
Tasks are classified, context is deliberate, tools are bounded, checks issue the verdict and evidence survives the run.
Course path
Nine focused lessons. One complete operating model.
The videos run for about 50 minutes in total. The deeper work is applying each lesson to the runnable companion repository or to a workflow in your own codebase.
- 01
From Software Engineer To Agentic Engineer
See the shift from using an assistant to designing a governed delivery system, and choose the production-adjacent workflow you will improve through the course.
3 min - 02
Decide What The Agent May Do
Separate autonomy from authority and turn task classification into inspectable policy instead of a vague prompt.
5 min - 03
Give The Agent Current Context And Durable Memory
Build finite context packs, choose retrieval paths and preserve the decisions that should survive the current run.
9 min - 04
Build The Delivery Loop
Define triggers, work selection, isolation, execution, verification, persistent state, retry rules and terminal outcomes.
8 min - 05
Make Verification Issue The Verdict
Use deterministic gates, fresh-context review and merge-readiness evidence to decide whether agentic output is engineering work.
7 min - 06
Route Work Across Models And Infrastructure
Define stable model lanes, route by task and data class, and measure whether the routing policy is actually economical.
5 min - 07
Split Work Across Controlled Roles
Separate orchestration, implementation, review and merge responsibility while keeping every handoff explicit.
4 min - 08
Earn An Evidence-Gated Merge Path
Assemble scope, policy, tests and product review into a merge path that stops unsafe or unsupported changes.
6 min - 09
Operate And Improve The System
Turn run evidence into a practical 30-day adoption plan and improve the workflow without mistaking activity for value.
3 min - A
Implementation Notes
Optional implementation detail for running and verifying the companion repository.
3 min · optional appendix
What you will be able to do
Leave with an engineering system you can explain and improve.
Classify delegated work by risk, authority and blast radius.
Build context packs and durable memory that agents can actually use.
Design bounded delivery loops with explicit stop conditions.
Make deterministic checks and evidence issue the engineering verdict.
Route work across local, hosted and specialist model lanes.
Split delivery across controlled roles without losing merge ownership.
Operate and improve an agentic workflow using evidence from real runs.
Start now
Becoming an Agentic Engineer
Pay once, register one named learner, and work through the course at your own pace. Your progress is saved so you can return whenever you need it.
£150per learner · one-off Stripe payment
Invited by your organisation?
Register for complimentary access.
If Radical Geek has approved your email address or organisation domain, register here. You still get your own named account and your own progress record.
Need this for a team?
Combine self-paced access with a private workshop.
Use the form to discuss team access, a live working session, or applying the operating model to one of your delivery workflows.