Principles Library / Task-fit delegation

Task-fit delegation

Automate by Task Nature, Not Capability

Act

Showing all 5 practices. Answer the Diagnose questions and the most relevant come first.

Classify every task before you automate it

Before automating, sort each task as rule-based, judgment-required and verifiable, or judgment-required and not verifiable, and record the call.

Start by asking your team: Is this task deterministic, or does it really need interpretation and judgment?

Avoid · You’ll know it’s working when

Avoid

Treating "AI can attempt this" as the same as "AI should do this"

You’ll know it’s working when

Tasks reach AI because they need judgment it can handle, and speed alone never decides it.

Send rule-based tasks to conventional automation

Route deterministic tasks to fixed code, not to AI.

Avoid · You’ll know it’s working when

Avoid

Running rule-based work through an LLM out of convenience or habit

You’ll know it’s working when

Rule-based tasks run as fixed code, and LLM calls are kept for work that needs them.

Write down what correct looks like first

Before handing a judgment task to AI, write down what a correct outcome looks like and how it'll be checked.

Start by asking your team: Can we say, before delegating, what a correct outcome looks like?

Avoid · You’ll know it’s working when

Avoid

Assuming a task is well-defined because it was easy to describe

You’ll know it’s working when

For every delegated task, someone can say what correct looks like and how it's checked.

Add a check before irreversible AI actions

Build a verification or review step into any AI-delegated task whose outcome can't be undone.

Start by asking your team: Who or what verifies this output before it takes effect?

Avoid · You’ll know it’s working when

Avoid

Delegating irreversible decisions because verification feels like it slows delivery

You’ll know it’s working when

Irreversible agent actions wait for a check that confirms the outcome is correct.

Revisit delegation decisions as tasks grow

Recheck delegation as tasks change. A task that was reversible and well-defined at small scale may not stay that way as scope grows.

Start by asking your team: If this task goes wrong, can we catch it and undo it before it compounds?

Avoid · You’ll know it’s working when

Avoid

Widening AI's task scope by reflex after early wins, without re-testing fit for the new tasks

You’ll know it’s working when

Each time an agent's scope grows, someone re-tests whether the task still fits.

Overview

Principle intent

Match the method of execution to the nature of the task, not to what AI is technically capable of attempting.

Read the full intent

Deterministic tasks, where a fixed rule reliably gives the right output, belong in conventional automation. Judgment tasks may suit AI, but only when they're well-defined, reversible and verifiable. Tasks that are ambiguous, carry irreversible consequence or depend on context the system can't access shouldn't go to AI, even if it can attempt them.

When it holds: what it does for you

Predictable rule-based work

Work a fixed rule can handle runs as fast, steady, auditable code.

Mistakes caught early

Irreversible AI actions get checked before they reach downstream systems.

Fair read on AI

When AI fails, you can tell whether the task was ever a good fit.

When it is ignored: from early signs to a lasting condition

Early warning signs · Visible now, cheap to fix

Rule-based tasks run through an LLM call when fixed code would do.
Tasks go to AI just because it can attempt them.
Ambiguous or high-stakes decisions go to AI because a person would take longer.

What happens over time · Accumulates, costly to reverse

Rule-based work is slower, less predictable and harder to audit than fixed code.
Irreversible mistakes show up only after they've spread through downstream systems.
Trust in AI erodes unevenly, blamed for failures on tasks it never suited.

The condition it becomes

Quality Fragility · Primary

Customers find defects before the team does.

How this principle feeds it

mistakes found after they spread are the late, costly defects this condition describes.

Oversight Erosion · Primary

The review process still runs. It stopped meaning anything.

How this principle feeds it

a check with no criteria still runs and passes, so it keeps the look of review without the substance.

Intent Drift · Primary

The team built exactly what was asked. It is no longer what is needed.

How this principle feeds it

when nobody can question the output, it keeps running against a misread or outdated intent.

Attribution Failure · Secondary

We know what happened. Nobody can explain why.

How this principle feeds it

failures get recorded against the AI while the delegation decision behind them goes unexamined.

Where is your system on this path?

Diagnose

AI signs are on by default. Use the AI switch to exclude them. This principle applies once agents produce or act on work in your delivery.

1. Which of these early warning signs are you seeing?

Rule-based tasks run through an LLM call when fixed code would do.
Tasks go to AI just because it can attempt them.
Ambiguous or high-stakes decisions go to AI because a person would take longer.
Nobody can say what correct looks like for a delegated task, or how it's checked.
Agents carry out irreversible actions before anyone verifies the outcome.
An agent's scope has grown past its original use case without a re-test.

2. Over the last few months, have you noticed…

For each consequence: Not seen / Sometimes / Persistently / Not sure.

Rule-based work is slower, less predictable and harder to audit than fixed code.
Irreversible mistakes show up only after they've spread through downstream systems.
Verification is bolted on afterwards to tasks never designed to be checked.
Judgment calls move into AI workflows where nobody is set up to question output.
Trust in AI erodes unevenly, blamed for failures on tasks it never suited.
AI takes on work by capability, faster than anyone can meaningfully review it.

3. The condition it becomes

Based on your answers, recognize at most two conditions: No / Partly / Yes, that is us.

Quality Fragility: Customers find defects before the team does.

Oversight Erosion: The review process still runs. It stopped meaning anything.

Intent Drift: The team built exactly what was asked. It is no longer what is needed.

Attribution Failure: We know what happened. Nobody can explain why.

Take your Diagnosis to the next level by invoking AI Diagnostic Coaching tool.

What else have you observed? Optional. Shared only if you choose to start coaching.

Start a Diagnosis session: review and edit a draft containing your note, conditions you recognized in Step 3 (including whether they match partly), warning signs and consequence ratings before sending it. No scores or principle files are sent.

See what practices will help you