Build agents that act for one employee, with authority, approvals and memory people will trust.
Jobs and autonomy, delegated access, approvals, prompt injection, memory, notifications, value and governance for agents that work for one employee, with a ten-worksheet toolkit for the agent PRD.
Guide PDFs and toolkits download with paid Premium while it is active, or with complimentary access we have approved. Days earned through referrals open the rest of the site, not these files.
Delegate / approve / remember / govern
Pages
80
Chapters
31
Diagrams
29
Dated sources
37
Worksheets
10
What is an enterprise personal agent, and what does the PM decide?
A personal agent reads one employee's mail and calendar, drafts their replies, chases their approvals and rebooks their travel. It acts with that person's access, inside rules their employer sets, and it writes in their name. Whether people hand it real work depends less on the model than on how much authority it has, what it asks before acting and what it remembers.
This guide follows one fictional company rolling out a personal agent to 6,000 employees across Europe, India and the US, from choosing the first three jobs to a works council review, a month-end incident and a cancelled flight at 1:40 a.m. You learn the decisions a PM owns and how to defend them to security, legal, employee representatives and an interviewer.
THREE COMPLETE PAGES, FREE TO READ
Judge the reasoning before you decide.
Each preview is a real page from the PDF. Open one to read it at full size.
Page 21
Fight approval fatigue with design
Why people stop reading approvals after about fifteen a day, and five techniques that bring attention back.
Page 29
Give the person a memory they can see and edit
What a personal agent may remember, what it never stores, and the memory page that makes it trusted.
Page 65
The day the field changed meaning
How a quiet ERP change sent 140 wrong chasers, and the tripwires a job needs before it acts on its own.
31 CHAPTERS / 6 PARTS
From the product decision to a design you can defend.
The product
Know what a personal agent is, read the platform landscape, choose jobs people will hand over and find the objections before the review does.
01Know what you are buildingp. 7
02Read the landscape before you buildp. 9
03Pick three jobs people will hand overp. 11
04Discover the work, the data and the objectionsp. 13
Authority
Act with less than the person's access, put a gatekeeper outside the model, design approvals people read, contain injection, register connectors and bound unattended work.
05Act as the person, with less than their accessp. 16
06Put a gatekeeper between the model and the worldp. 18
07Design approvals people read before they tapp. 20
08Treat prompt injection as a question of data flowp. 22
09Connect systems through a register, one connector at a timep. 24
10Keep work going while the person is awayp. 26
Memory and personal context
Give people a memory they can see and edit, personalise without surprise, assemble context by rule and plan for role changes and leavers.
11Give the person a memory they can see and editp. 29
12Personalise without surprising the personp. 31
13Assemble context the same way every timep. 33
14Plan for people who change roles or leavep. 35
Where it lives
Interrupt only when it is worth it, stay one assistant across surfaces, design out misuse with care for the person and keep data in its region.
15Interrupt only when it is worth itp. 38
16Meet people where they already workp. 40
17Design out misuse, and care for the personp. 42
18Keep each person's data where it belongsp. 44
Run it
Price per person and per job, measure value from system data, evaluate the jobs, keep an audit, pass governance, roll out in waves and operate a service.
19Price the agent per person and per jobp. 47
20Measure value people and finance both believep. 49
21Evaluate the jobs people hand overp. 51
22Observe the service and keep an audit people can defendp. 53
23Get governance to a signaturep. 55
24Roll out in waves that teach you somethingp. 57
25Operate it as a service people depend onp. 59
Prove and defend
Work through three cases, write the agent PRD, defend the design in a clinic and plan your first six weeks.
26Case: the account manager's meeting briefsp. 62
27Case: the finance analyst's month-end chasersp. 64
28Case: the field engineer's cancelled flightp. 66
29Write the agent PRDp. 68
30Design clinic: defend the personal agentp. 70
31Your first six weeks as the PMp. 72
WORKED INCIDENTS
Three cases, worked from the evidence.
Each fictional case follows one job from design to its first serious failure, and asks what a PM has to answer: what the agent assumed, who noticed, how it was contained and what changed.
Ambiguous data / visible assumption
The brief that mixed two customers
Resolve accounts from email domains before names, and put the riskiest assumption at the top of the brief.
Act and report / silent ERP change
The day the field changed meaning
Add contract tests on field meanings, automatic pause rules and a way for any user to stop a job for everyone.
Lost reply / new request identifier
The rebooking that happened twice
Build request identifiers from stable fields, show unknown outcomes as unknown and reconcile bookings daily.
INCLUDED / 14 FILES
Ten worksheets for the agent PRD.
The guide ships with ten editable worksheets and three spreadsheets that become the sections of an agent PRD and the review packet security, data protection, legal and employee representatives sign.
Start with the job and autonomy map and the delegation matrix. Together they answer the two questions every reviewer asks first: what will it do on its own, and with whose access?
These questions, concepts and lessons pick up where the chapters leave off. Each link also sits, clickable, in the chapter it supports, and keeps its own access level on the site.
An AI agent that works on behalf of one employee: reading their mail and calendar, drafting their messages, chasing their approvals and handling tasks such as travel changes, with that person's narrowed access and inside rules their employer sets.
Do I need to be technical to use this guide?
No. It is written for product managers and explains delegation, approvals, injection and memory through the decisions you own. It shows a short policy rule and an approval card so you can recognise them in a review, but never asks you to configure anything.
How is this different from a chatbot or a copilot?
A shared assistant answers anyone with the company's voice. A personal agent acts as one person, with their access, and its mistakes land on their name. That changes how you design authority, approvals, memory and measurement, and which stakeholders you need on side.
Which platforms and rules does it cover?
Microsoft 365 and Agent 365, Google's Gemini Enterprise Agent Platform, AWS Bedrock AgentCore, OpenAI's Agents SDK, Anthropic memory controls, MCP and A2A, the EU AI Act and digital omnibus, India's DPDP Rules and the NIST AI RMF, with 37 sources checked in October 2026.
How current is it?
Platform, protocol and regulatory facts are dated October 2026 and several cited standards are drafts. Each chapter teaches the durable product decision first, so the guide stays useful when products change.
What is in the toolkit?
14 files: ten editable worksheets from the job and autonomy map to the incident review, plus CSV versions of the delegation matrix, the connector register and the value plan. Each has one worked example from the guide.
Will this help in AI PM interviews?
Yes. Chapter 30 is a design clinic of ten prompts with a scoring rubric and worked answers, and every chapter links to related practice questions on the site. Email agents, travel agents and approval design are common AI PM interview cases.
Is it included in Premium?
Yes. The full PDF and toolkit come with paid Premium while it is active, or with complimentary access we have approved, and there is nothing extra to buy. Anyone can read three full sample pages first. Days earned through referrals open the rest of the site, not the guide files.
How was it produced?
AI tools helped with research, source checks, drafting and the cover art. The diagrams are vector figures written in code. Maridian Group, Helm and every person in its scenarios are invented. This edition is ready for publisher review, and automated or AI review is no substitute for an independent human editor signing it off.
