Before this level
L2 first
In practice most teams do L2 first, because L2 gives them the ground this level builds on. If your team already has it, the assessment will say so and we will start you here.
Level 3 of three · 24 hours · online, own pace
L3 — Run and Guard It
The agent goes into daily use. Your staff learn to watch it, correct it, keep it inside the rules, and keep records that stand up to a question.
L3 is level 3 of the OwnYourAI corporate AI training path. It is 24 hours of online, own-pace learning delivered as 12 online sessions, and it is the point where the agent your team built goes into daily use, watched and guarded.
L3 · 12 online sessions · part of a 48-hour path
24 hours12 online sessions
Own paceonline
Level 3 of threeof a 48-hour path
On your worknot textbook examples
What this level is for
Where it sits in the path
After this level
Choose whether to continue
You are not obliged to continue. Plenty of companies stop here, run with it for a quarter, and come back when they need the next piece.
Session by session
The 12 sessions, one at a time
Every session is 2 hours of online learning at the person's own pace, with a scheduled live touchpoint inside it and a trainer present. Each one ends with something your company keeps: 24 hours in total for L3.
| # | Session | The one thing it settles | What your company keeps |
|---|---|---|---|
| 1 | Handing it over into daily use | What "live" has to mean before the agent touches real work | A signed go-live checklist: who owns it, what it may touch, the fallback when it is unavailable |
| 2 | The weekly watch | What the log must show, and who reads it | The watch log format, with the reader's name and the first week's entries filled in |
| 3 | Catching it going wrong | The correction loop, and how a fix is never lost | A correction log with version history for the agent's instructions |
| 4 | Data in daily use | What may go in and what never does, re-tested against what it is now asked to do | The live data card and a re-dated do-not-enter list |
| 5 | The rules that bind the company | Which obligations actually apply, and who is accountable for each | An obligations register with the accountable name, and open items marked open |
| 6 | The policy a third party can read | What management has decided and signed | The approved AI use policy v1, with an approval date and a review date |
| 7 | Evaluating it properly | What "good enough" means as a score rather than a feeling | An evaluation scorecard on twenty fresh inputs, with the pass line the company set |
| 8 | Drift: what changes underneath you | How to tell whether the agent quietly got worse | A dated re-test procedure and its calendar |
| 9 | Records that stand up | What an evidence pack contains, and who may read it | The evidence pack, filled once, answerable by a single named date |
| 10 | Escalation | What always needs a person, and which person | The escalation matrix, with named roles and a first-response time |
| 11 | The incident drill, live | What the company does in the first hour of a real failure | The incident report from the drill, and the corrected rule that came out of it |
| 12 | The run-and-guard assessment | Whether your company can actually run and defend its agent | A live assessment on fresh inputs, and your run-and-guard record |
The standard every run is held to
The three things a guarded agent always has
A company that cannot produce these three has an agent running on trust. They are written in the sessions, by the people who operate it, in the company's own words.
Someone reads the log
Every week, named
The watch log has a reader with a name against it. Fifteen minutes on a Monday, written down, including the weeks where nothing went wrong — an empty log and an unwatched agent look identical.
Every failure has a written correction
And a version behind it
The confident wrong answer, the right answer with the wrong detail, and the output that widens its own scope each get a written fix, stamped so the company can say which version produced which output.
The record is answerable by date
Not reconstructed from memory
What the agent did on a named day, under whose authority, and who signed it. The evidence pack is assembled once, for real, and read back at the assessment.
The assessment
How L3 is marked
The assessment is delivered live online with a trainer present. It is not a multiple-choice quiz, and it does not test whether someone remembers AI theory: it tests whether they can do the thing on their own desk.
| Part | What the person does | Time | Marked by |
|---|---|---|---|
| A. The agent under load | Runs the agent on two inputs the assessor supplies fresh at the assessment, and states what it may do alone and what it may never do | 20 min | Trainer |
| B. The scorecard | Explains the evaluation score, the pass line, and what would take the agent out of daily use | 10 min | Trainer |
| C. The record | Produces the evidence pack for a single named date, and says who may read it | 10 min | Trainer |
| D. The incident | Is given a described failure and gives the containment step, the correction, and the notification decision | 10 min | Trainer |
A. must produce a usable output on the inputs supplied fresh at the assessment: a prepared demonstration does not pass. B., C. and D. must each be answered correctly. Anyone who misses a part is assessed again on that part at no extra cost, and that re-assessment is recorded as such. The assessor is named on the record, with the date.
What is written down about it
Per person, kept together and retrievable on its own: the identity record, attendance at each session, progress at each session with the work produced, the assessment result part by part, the artefact that was demonstrated, and the name of the assessor with the date. Everything on that list is written by a named trainer, not by the learner. What we keep, and who can read it is set out on its own page.
Written to the SSG funding requirements as published. Course funding application status: pending.
What your team can do afterwards
The capability this level leaves behind
These are the skills that travel between companies, which is why the level is written around them. About 70% of every level is that shared skill; about 30% is your own workflows.
- Artificial Intelligence Ethics and Governance
- Generative AI Model Evaluation
- Automation Management
The shape of L3
| What | Detail |
|---|---|
| Level | L3 of three — Run and Guard It |
| Hours | 24 hours |
| Delivered as | 12 online sessions, online and own pace |
| Scheduled touchpoints | Yes — online, booked in advance inside the level |
| Working language | Plain English. No technical background needed. |
| What you keep | The work produced, the written record, and the skill in your own staff. |
What L3 is sold as
A result, how it is measured, when it lands, and what happens if it does not.
- The result
- The agent goes into daily use. Your staff learn to watch it, correct it, keep it inside the rules, and keep records that stand up to a question.
- How it is measured
- On the work your team produces and can run without us: the workflow, the policy, the agent or the operating routine that this level is built to produce — checked against your own real work, not a model answer.
- When it lands
- Inside the weeks agreed at enrolment, at your team's own pace.
- If it does not land
- If the level has not produced what it promised, your team stays on it with us at no extra cost until it has, or the seat is not charged. We would rather be fair than clever about that.
L3 — questions
What companies ask about this level
How long is L3 and how is it delivered?
L3 is 24 hours of online learning in 12 online sessions. It is own-pace, so your staff choose the hours, and every level has scheduled online touchpoints so nobody is left alone with it.
Does L3 need any technical background?
No. The levels are built for people who do the work, not for programmers. If your staff can use a spreadsheet and follow a process, they can do this. Where a level needs something technical, we show it step by step and you can ask us inside the touchpoints.
Can we start at a later level if our team already uses AI?
Yes, after the build assessment. The assessment is how we agree the right starting point, so you do not pay for hours your team does not need. We would rather place you correctly than sell you a level.
What do we get at the end of L3?
A documented outcome, not a certificate alone: what your team can now do, the work they built, and the records that go with it. If a release is not something you are satisfied with, we fix it before you accept it - that seems fair.
Previous: L2 — Build Your Company's Own AI Agent ·
Next step
Book the build assessment
The assessment tells you whether L3 is the right place to start, and what it would cost for the seats you actually need.
Corporate pricing on request. Course funding application status: pending.