Guide10
Vulnerability checks that actually catch things.
AI scanners plus human judgement — what we look for before a release goes live.
12 Apr 2026·3 min read
09Resources · Workshop
Shipping with agents without silent debt.
A practical session for teams already running agents, where the constraint has stopped being how fast code appears and become how fast it can be trusted.
Workshop30 Apr 20262 min readFrom £295
- Format
- Remote, or on site in the UK with travel at cost
- Duration
- Half a day (3.5 hours)
- Group size
- Up to 8, or up to 16 with a second facilitator
- You leave with
- Written guardrails and two metrics wired up
01The session
Control, not capability.
This assumes you are already using agents daily, so we spend the time on keeping the output reviewable rather than on getting more of it.
What we cover
- 01Scoping tasks so an agent returns a reviewable diff instead of a pull request nobody opens
- 02Keeping changes small on purpose, and why large agent output should be rejected rather than read
- 03Deciding which categories of change an agent may never own
- 04Tests and checks that must pass before a human even looks
- 05Naming a human owner per change, and what ownership actually obliges them to do
- 06Reading agent output for the specific ways it fails, rather than skimming it like a colleague’s work
Who it is for
- Teams running agents daily with a review queue growing faster than they can read it
- Leads who suspect approval has quietly replaced review
- Engineers inheriting code no human ever really accepted
- Anyone who cannot say who decided the thing that broke
02
The problem with volume.
Agents change the shape of the bottleneck. Producing a change is close to free; understanding one is not, and understanding has not got any cheaper. Teams that do not adjust end up with a review queue growing faster than anyone can read it.
What follows is predictable. Review becomes approval. Approval becomes a formality. Six weeks later something breaks and nobody can say who decided the thing that broke it. That is what we mean by silent debt: not bad code exactly, but code no human ever really accepted.
03
Guardrails we recommend.
The set that survives contact with a real backlog is short. Every change has one named human owner. Anything touching data shape, authorisation or money is human-authored. Diffs above a size the team agrees on get split rather than approved. And no change merges on the strength of a model reviewing another model.
None of this is exotic. It is mostly the discipline teams already claimed to have, made explicit because the volume no longer forgives informality.
04
Worth measuring.
Two numbers tell you whether the process is real. How long a change waits for review, and what proportion get sent back. If the wait is growing, the queue is winning. If almost nothing is ever returned, review has quietly become a rubber stamp.
We work in your repository against your current backlog, and you leave with the guardrails written down and both metrics wired up somewhere the team will actually see them.
03Pricing
Book a seat, or the room.
For teams already using agents who need the review culture to catch up.
Public seat
£295
per person
One place on a scheduled open session.
- Live half-day session
- Review checklists you keep
- Recording and notes afterwards
Most chosen
Private team
From £1,450
per half-day, up to 8 people
Run against your repository, your agents and your merge process.
- Audit of how agents are used today
- Review checklist tuned to your stack
- Worked examples on your own changes
- Written follow-up and agreed next steps
Private team, larger group
From £1,950
per half-day, 9 to 16 people
A second facilitator so breakouts stay useful.
- Everything in Private team
- Second facilitator in the room
- Breakout exercises in small groups
Rescheduling is free with 5 working days’ notice. All prices exclude VAT.
