Jobs-to-Be-Done Agent Mapping
Bring order to AI chaos by listing every stakeholder's jobs-to-be-done, then mapping agents onto them.
- Difficulty
- Moderate
- Time to result
- ~months to results
- Steps
- 4
- Confidence
- 88%
A method for deploying AI/agents across a company without descending into scattered experimentation. Start from the jobs-to-be-done for each stakeholder (users, advertisers), enumerate them concretely, then identify where agents fit, build cross-functional teams around those jobs, and track progress against the business outcome of each job. Use it when AI enthusiasm is producing lots of activity but little focus.
Origin
Spiegel describes wanting to bring 'order to that chaos' as everyone at Snap built agents. The mechanism was to get 'really really basic' about jobs-to-be-done — e.g. for Snapchatters: get people to download, add close friends, use lenses; for advertisers: onboard, configure a campaign — and let that list reveal where agents belonged.
Core principles
- 01Let a thousand flowers bloom on AI, but anchor it to what matters to customers.
- 02Define jobs-to-be-done concretely for each stakeholder group.
- 03The job list reveals where agents fit and where to build focused cross-functional teams.
- 04Each job doubles as a tracking mechanism against a business outcome.
- 05If you can define a job clearly enough, you can often build an agent to do the whole workflow.
How to run it
- 1
Enumerate jobs-to-be-done per stakeholder
Get basic and specific. List the concrete jobs for each group — e.g. for users: download the app, add close friends, use lenses; for advertisers: enter the ad platform, configure a campaign.
Pro tip Keep it 'really really basic and straightforward' — resist abstraction.
- 2
Map agents onto the jobs
With the jobs listed, it becomes clear where agents can help and where you need focused, AI-supported cross-functional teams built around specific jobs.
- 3
Track outcomes per job
Use the job list as the mechanism to track progress against the business outcome each job is meant to drive, keeping AI work tied to what matters.
Watch out Without this anchor, agent-building becomes scattered experimentation disconnected from customer and business value.
- 4
Automate whole workflows where the job is clear
Where a job is defined clearly enough, build an agent that executes the entire workflow end to end — e.g. a go-to-market agent that takes a product idea, writes the spec, identifies sign-off owners, runs legal/trust-and-safety risk analysis, and drafts the launch materials.
Pro tip Doing it 'in one shot' across the workflow creates the biggest lift.
In the wild
Snap built an agent that takes a product idea and writes the spec, identifies the people needed for sign-offs, does the legal/trust-and-safety risk analysis, and writes the go-to-market materials like the blog — moving toward doing it in one shot.
→ A whole cross-functional workflow compressed into a single agent-driven pass.
On the internal app build, users shake to report a problem and agents debug exactly what went wrong and suggest a fix, with automated code review that Spiegel says has caught close to 10,000 bugs.
→ More people can safely submit code at near-billion-user scale without breaking things.
Common mistakes
Building agents without an anchor
Unbounded AI experimentation produces chaos; without tying agents to concrete jobs-to-be-done, effort scatters away from customer and business outcomes.
Defining jobs too abstractly
Vague jobs can't be mapped to agents or tracked; the value comes from getting basic and specific about each concrete job.
Is it for you?
Best for
Leaders rolling out AI/agents across a company amid scattered, unfocused experimentation.
Not ideal for
Early exploratory research where the goal is open-ended discovery rather than workflow automation.
From the transcript
“we really wanted to bring some like order to that chaos”
“by listing out all these jobs to be done, you know, really for the for the community journey and for advertisers as well, it became…”
“if we can uh define a job to be done clearly enough that we can build an agent to do it”
From the episode
Snapchat CEO: Why distribution has become the most important moat
Evan Spiegel