AT&T Israel · AI for Product Managers

One context. Every tool. One standard.

~40PMs
2Cohorts · ≤20
2 × 2hZoom
2 × 5hHalf-days
The challenge

Two problems

01

Context is lost between tools

6 tools · 17 AI assistants and agents · 6 separate memories · 1 PM carrying the context.

✕ ✕ ✕ ✕ ✕ ✕ re-explain copy-paste re-explain re-explain paste by hand re-explain PM carries the context in their head Teams · Outlook M365 Copilot Chat Facilitator · recap Researcher ◍ OWN MEMORY Word · Excel · PPT Copilot in Word Copilot in Excel/PPT Copilot Notebooks ◍ OWN MEMORY Internal AI Ask AT&T Discovery Partner Requirements Gatherer ◍ OWN MEMORY VS Code GitHub Copilot Own skills Own agents ◍ OWN MEMORY Jira · Jira Align Rovo in the backlog Jira Align (Rovo beta) ◍ OWN MEMORY Rovo Rovo Chat Rovo agents Rovo Search ◍ OWN MEMORY

Contractors: no M365 Copilot — Teams and Office work happens without AI

TodayContext lives in each tool's chat
GoalContext lives in shared files every tool reads
02

No shared standard

Every department builds its own way. Nobody can say what "good" looks like.

?

Department A

PRD template · v3
Problem
Scope
Acceptance criteria
SDD flow "A"3 own agents
?

Department B

Feature brief
Overview
User stories
Same SDD, other name5 own agents
?

Department C

Requirements doc
Brief
Requirements
Metrics · Risks
SDD flow "C"Prompts, no agents
?

Industry standard — missing. No shared benchmark to build against or check quality with.

Today"I built skills and agents — but are they good?"
GoalOne standard to build and check against
The approach

Shared files. One standard.

01

Communicate with files

Every tool reads the same files first — and writes back when done.

Initiative folder CONTEXT · STATE · decisions spec · sources · handoffs READ FIRST · WRITE BACK Teams M365 Copilot Word · Excel VS Code Agents · skills Jira · Rovo
02

The industry standard

Google & Anthropic practice, for non-coders in VS Code and Copilot.

1

Define the job

One job per agent · clear trigger · job-brief instructions

2

Equip it

Examples · only the tools it needs · small context · files as memory

3

Structure it

Small steps · human gates · maker vs checker · rules vs judgment

4

Trust it

Test like a product · guardrails · version it

Learn before we commit

Zoom first. Workshops built on what we hear.

STEP 01

Readiness

Access · project

STEP 02

Zoom

Standard · files

GATE

Listen & refine

Adapt to their day-to-day

STEP 03

Half-day 1

5h · own project

STEP 04

Practice

Real work

STEP 05

Half-day 2

5h · own project

STEP 06

Clinic

1h review

We learnSo the half-days get
Where context breaksLabs that follow their real tool path
Their projectsHands-on work on their own initiative
Their real levelThe right cohort and lab depth
Access gapsFixed before day 1
The sessions

Eight sessions, two tracks

Click a session to see what it covers.

01Zoom A · Exploration ZoomIntermediate2h
  • How leading teams work with AI — the industry standard in plain terms
  • Why context gets lost between tools, and the fix: files as shared memory (live demo)
  • What a skill is, and when to use one instead of a prompt
  • Breaking a complex task into small steps — meeting → decisions → spec → stories → Jira
  • Discussion: where do you lose context today? (feeds the workshop design)
02Zoom B · Exploration ZoomAdvanced2h
  • The standard for building agents: one job, clear trigger, job-brief instructions, minimal tools
  • Patterns: steps with human gates, maker vs checker with a stop rule, testing your agent
  • First look at agents that communicate through files (writer → critic → publisher demo)
  • Live review: volunteers' own agents checked against the standard
  • Capture projects, level and access gaps for the half-days
03Half-day 1 · Set up and break it down In personIntermediate5h
  • Create an initiative folder for your own project; convert sources to markdown
  • M365 Copilot: meeting → decisions, Researcher / Notebook brief written back to files
  • Break your task into steps; run a provided skill for each step in VS Code
  • Apply it to your own project + write-back ritual (STATE, decisions)
  • Practice assignment for the week
04Half-day 1 · Build to the standard In personAdvanced5h
  • Layered instructions + initiative folder for your project
  • Turn a repeated task into a prompt file and a skill with an example
  • Build a custom agent with a clear description and only the tools it needs
  • Test it: 3 golden inputs + should / shouldn't-trigger checks
  • Apply it to your own project
05Half-day 2 · From project to Jira In personIntermediate5h
  • Show & tell from the practice week
  • Write stories with provided skills → critic agent → fix
  • Atlassian MCP: dry-run → approve → create epic and stories in sandbox Jira
  • Internal speakers: AT&T Jira setup and internal agents (FYI)
  • Write your first own skill · personal "which tool, when" card
06Half-day 2 · Agents that communicate with files In personAdvanced5h
  • Show & tell from the practice week
  • Chain writer → critic → publisher using hand-off files and human gates
  • Critic loop with a stop rule; fixed rules as a checklist, judgment for the AI
  • Publish to Jira via MCP with dry-run and approval
  • Agent design canvas for one real AT&T workflow · version your kit
07Clinic ZoomIntermediate1h
  • Bring a real workflow — reviewed live
  • Fix what's blocking adoption
  • Share wins in the Teams channel
08Clinic ZoomAdvanced1h
  • Your skills and agents reviewed against the standard
  • Next improvements and what to share with others
  • 30-day check briefing
Intermediate leaves with

Context across tools

  • Initiative folder
  • A task broken into steps
  • Stories in Jira — no copy-paste
  • First own skill
Advanced leaves with

Agents to the standard

  • Tested skill + custom agent
  • Writer → critic → publisher via files
  • Agent design canvas
  • Versioned personal kit
→

Next: confirm cohorts, dates, Copilot Business seats, Atlassian MCP + sandbox Jira, room and co-facilitator.