Role-to-Task AI Decomposition
Break a job into tasks before deciding what AI should automate or assist
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 6
- Confidence
- 99%
Role-to-Task AI Decomposition treats a job title as an imprecise container and the actual work as a sequence of discrete tasks. A team selects one role, lists its recurring activities, and defines the inputs, decisions, outputs, tools, and risks associated with each activity. Every task is then scored for current AI capability, data availability, reliability requirements, and the importance of human judgment or relationship. Suitable tasks may be automated; uncertain tasks become human-reviewed; sensitive tasks remain human-led. The assignments are assembled into a workflow with explicit handoffs and monitoring. As models improve, the team can reassess individual tasks without making the misleading claim that an entire job is either automated or untouched.
Origin
Extracted from Marketing Against The Grain, where the hosts decomposed business development work into company research, fit identification, data gathering, personalized email, calls, decision-maker mapping, and social outreach.
Core principles
- 01Treat jobs as bundles of tasks
- 02Evaluate each task independently for AI readiness
- 03Preserve human ownership where judgment or trust matters
- 04Reassemble suitable tasks into an orchestrated workflow
- 05Expect modality readiness to change over time
How to run it
- 1
Select and observe the role
Choose one role and gather evidence about what its practitioners repeatedly do. Use real workflows rather than the formal job description alone.
Pro tip Review calendars, systems, and completed work products.
Watch out Job descriptions often omit coordination and exception handling.
- 2
Create the task inventory
Break the role into discrete actions with identifiable triggers and outputs. Separate research, judgment, communication, and administration.
Pro tip Use verb-led task names such as research accounts or map decision makers.
Watch out Tasks that remain too broad cannot be assessed accurately.
- 3
Define task contracts
For each task, specify required data, tools, quality criteria, risks, and downstream consumers. Note where tacit context is essential.
Pro tip Include examples of acceptable and unacceptable outputs.
Watch out Automation without an explicit quality threshold is difficult to govern.
- 4
Score ownership
Assess whether the task should be AI-led, human-led, or hybrid based on capability, reliability, sensitivity, and economics.
Pro tip Distinguish technical possibility from socially acceptable use.
Watch out Do not automate a channel simply because a model can generate content for it.
- 5
Reassemble the workflow
Connect assigned tasks into a sequence and define review, escalation, and handoff points. Preserve context between stages.
Pro tip Give the human reviewer the evidence behind the AI output.
Watch out Individually successful automations can still fail when their interfaces do not align.
- 6
Reassess over time
Track performance by task and revisit assignments as capabilities, costs, and risks change. Expand only where evidence supports it.
Pro tip Maintain a capability matrix rather than rewriting the entire role design.
Watch out Do not assume every task will progress toward full automation.
In the wild
A company divides the BDR role into account prioritization, research, contact mapping, email drafting, calling, and social outreach. AI prioritizes accounts and drafts individualized emails; humans review outreach, handle calls, and retain control of social interactions until quality and channel norms justify changes.
→ The company gains scalable assistance without making an all-or-nothing decision about replacing the role.
Common mistakes
Assessing the whole job at once
A role combines tasks with very different data, risk, and relationship requirements, so a single automation verdict is misleading.
Ignoring channel acceptability
Technically feasible automation may be harmful when it floods personal or social channels and erodes trust.
Forgetting orchestration
Automating isolated tasks does not create value if context and outputs fail at the handoffs between them.
Is it for you?
Best for
It is best for teams redesigning knowledge-work roles or building AI agents around established workflows.
Not ideal for
It is not ideal for work that is too irregular or tacit to describe as observable tasks and outputs.
From the transcript
“if you think about every role when you think about AI, you should just break it down into a list of tasks. That's the best…”
“That that task is how do I research companies? How do I find good fit companies? Then how do I get enough data about them…”
“I don't ever want to see social outreach because I don't like thinking of AI mass saturating people's social channels. I think that one should…”
From the episode
Steal HubSpot’s Go To Market Strategy With Ai For 2025