Implementation Notes

3A Before You Build an Agent

3 min read · Implementation Notes · Updated Oct 2026

3A Before You Build an Agent
80/15/5 is a planning heuristic, not a measured mix and not an ROI claim.

Assign each process before anyone builds. A form can run on rules, a concession email stays with a named person, and a refund runs alone only under a cap you wrote down.

Part 3 — gated agent practice. Part 1 is click-and-do pedagogy: Click-and-Do Before Slide Decks. Part 2 is role-fit on the analysis path: Role Paths Before Generic Analytics.

The figure’s 80/15/5 split is a planning heuristic, not a measured mix and not an ROI claim. Assign each process a band before anyone builds. Control maturity is a different axis, in From Prompt to Agent.

Three processes, three bands

Band Trigger Why this band
Automate A form arrives with an email, and the domain is not free mail. Rules fire. No one judges the row. The miss is calling that flow an agent.
Assist A customer asks for a concession. No cap is written down. A named person checks the amount, then sends.
Autonomize A refund request, the order already delivered, and the amount at or under a cap the team wrote down. That cap is a threshold you set, not a Prompt Anatomy price. Under the cap, write an audit row and refund. Over the cap, escalate.

Northline’s enablement lead (composite) shipped the concession email as Autonomize. The credit in the mail was fabricated. The name on the project was not the failure. Skipping the confirm was. A written cap plus an audit row is what moves a refund to Autonomize. The concession had neither.

A research job that also sends is the same miss. Split find from send in A Shortlist, Not Another Inbox. A ticket-type split between the concession email and the refund is a router, not a third agent. Bounded tasks and escalation live in How to Design an AI Agent Workflow.

Write three of your processes and assign a band. If all three land on Autonomize, the list is hype.

Finishing the Agent path shows judgment inside the drills. The system you run in production still belongs to you. For how this site and the training host fit, start with The Prompt Anatomy Ecosystem Map. The drills are not the runtime: The Model Is Not the System.

Who can test M10-12

We are recruiting a limited tester cohort while we close remaining bugs, check tools and schemas, and finish design polish. Soft launch is not a promise that every interactive schema passed full manual QA. There is no Stripe Corporate plan.

Who: Existing Core buyers with Modules 1–6. See plans and pricing. M7–9 helps; it is not required. Starter (Modules 1–3) is out until Core.

Offer: Free M10–12 access during the soft-launch window. This is a tester grant, not a public price change. 399 EUR is a B2B IVS workshop (3 hours, ≤10 people, license excluded)—not a Stripe plan and not what checkout sells today.

Apply: Email info@promptanatomy.app with subject M10-12 tester, your purchase email, and LT or EN.

Ask: Work the path within 30 days. Reply with three bullets—what confused you, what worked, one bug or typo—plus one sentence on which process you put on Autonomize and why it is not Assist. Feedback shapes the general-availability cut. It does not replace support SLAs.

On this page

Move from pilot to program

Structured training for teams implementing AI under real operational and compliance constraints.

Explore training

Continue learning

Step 5 of 31 in Implementation Notes · Full reading order

Cluster hub

Implementation Notes

The Prompt Anatomy Ecosystem Map

Where to read, practice, and build—promptanatomy.blog for frameworks, promptanatomy.site for discover-and-try, promptanatomy.app for training, and how other properties fit without duplicating content.

9 min read · Implementation Notes · Updated Oct 2026

Go deeper

Framework

The Model Is Not the System

Teams fail when chat is the product. The model is the engine; context, memory, human judgment, tools, workflow, and output create the advantage.

8 min read · Framework · Updated Nov 2026

Template

Templates

Governance RACI Worksheet

Copy-paste RACI worksheet to assign accountable owners for AI workflow changes, releases, and incidents.

3 min read · Templates · Jun 2026

FAQ

Who can test Modules 10-12?

Existing Core buyers with Modules 1-6 on promptanatomy.app. The Data Analysis path (M7-9) helps but is not required. Starter buyers need Core first. Email info@promptanatomy.app with subject M10-12 tester, your purchase email, and LT or EN.

Can I buy a Corporate Stripe plan today?

No. 399 EUR is a B2B IVS workshop (3 hours, up to 10 people, license excluded), not a Stripe plan. Live checkout is Starter and Core only. M10-12 remains a tester grant for existing buyers.

Does finishing the Agent path replace production agents or governance?

No. Practice inside the drills proves judgment. Workflow IDs, eval gates, RACI, and the runtime still belong on your systems.

Must I finish Modules 7-9 before M10-12?

No. Modules 1-6 are enough to join the tester cohort. M7-9 is an operator grant for Core buyers, not a checkout SKU and not a gate for this grant.