Who does what: a 2x2 for every AI task

A Framework for Deciding

You have mapped your problem. You have evaluated the tools available. Now comes the decision: who does what? This is where the Delegation Matrix comes in.

The Delegation Matrix is a simple 2×2 framework that helps you make structured decisions about whether a task is best handled by you, by AI, by both of you together, or whether it should be reconsidered entirely. It works by assessing two dimensions:

• Human Strength: How much does this task benefit from human judgement, creativity, empathy, contextual understanding, or accountability?
• AI Strength: How much does this task benefit from speed, pattern recognition, consistency, scale, or information processing?

Every task you face can be placed somewhere on this matrix.

Quadrant 1: High Human Strength, Low AI Strength — Keep for Yourself

These are tasks where your human capabilities are essential and AI adds little value. They require judgement, emotional intelligence, ethical reasoning, or deep contextual understanding that AI simply cannot provide.

Examples:

• Making a difficult personnel decision. You are deciding whether to promote someone. This requires understanding their work, their growth trajectory, team dynamics, and organisational needs. AI has none of this context, and even if you described it all, the accountability for the decision rests with you.
• Navigating a sensitive client conversation. A long-standing client is unhappy. The way you listen, empathise, and respond in real time — reading tone, adjusting your approach — is fundamentally human work.
• Setting the strategic direction for your team. Strategy requires weighing competing priorities, understanding your people, reading the market, and making bets. AI can inform these decisions, but it cannot make them for you.
• Giving someone honest, caring feedback. The words matter, but so does the relationship, the timing, and the way you deliver them. This is not a task to outsource.

The rule for Quadrant 1: do it yourself. Trust your human capabilities. These are the tasks where you add the most irreplaceable value.

Quadrant 2: High Human Strength, High AI Strength — Collaborate

These are the most exciting tasks on the matrix. Both you and AI bring something valuable to the table. The best results come from working together — iterating, challenging each other, combining your respective strengths.

Examples:

• Writing a thought-leadership article. You have the original ideas, the voice, and the perspective. AI can help you structure your argument, generate alternative framings, suggest evidence, and produce a first draft that you then reshape and refine.
• Conducting research synthesis. You know what questions matter and how to evaluate source quality. AI can process large volumes of material quickly, surface relevant findings, and identify patterns across sources that would take you hours to spot.
• Brainstorming solutions to a complex problem. You understand the constraints, the stakeholders, and the politics. AI can generate a wider range of options than you might think of alone, including ideas from adjacent domains that you might not have considered.
• Designing a new training programme. You know your audience, their needs, and what works in your context. AI can help you structure content, generate examples, create exercises, and draft materials that you then adapt and improve.

The rule for Quadrant 2: collaborate actively. Use AI as a thinking partner. Bring your expertise and judgement to bear on everything it produces. This is where the combination of human and AI capability produces genuinely better outcomes than either alone.

Quadrant 3: Low Human Strength, High AI Strength — Automate

These are tasks where AI can do the work faster, more consistently, and often better than you can — and where the human judgement required is minimal. They tend to be well-defined, repetitive, or procedural.

Examples:

• Reformatting a report from one structure to another. You know what the output should look like. AI can do the actual conversion in seconds rather than the thirty minutes it would take you.
• Summarising meeting notes. The meeting happened. The notes exist. Extracting the key decisions and action items is straightforward processing work that AI handles well.
• Generating first drafts of routine communications. Standard project updates, status reports, acknowledgement emails — these follow predictable patterns that AI can produce reliably.
• Cleaning and organising data. Removing duplicates, standardising formats, categorising entries — these are tasks where human attention adds little value but human error is a real risk.
• Translating content between languages. For most professional contexts, AI translation is now good enough for internal use, with light human review for anything external-facing.

The rule for Quadrant 3: automate confidently, but verify. Set up the task clearly, let AI execute, and do a quick quality check on the output. Do not spend time doing manually what AI can do in seconds.

Quadrant 4: Low Human Strength, Low AI Strength — Rethink

This is the quadrant people forget about, and it is arguably the most valuable. If neither you nor AI is well-suited to a task, that is a signal worth paying attention to.

Examples:

• Writing a weekly status report that nobody reads. If the task adds no value, the answer is not to make it faster with AI. The answer is to stop doing it, or to replace it with something that does add value.
• Manually tracking information that a system should be tracking automatically. Perhaps the task exists because of a broken process, and the real solution is to fix the process.
• Creating a detailed plan for a project where the requirements keep changing. If the situation is too uncertain for detailed planning, both you and AI will produce plans that become obsolete immediately. Perhaps a lighter, more adaptive approach is called for.
• Compiling a report by copying and pasting from multiple sources. If this task is tedious for you and error-prone for AI, it might be a sign that the underlying systems need better integration.

The rule for Quadrant 4: question the task itself. Ask why it exists. Ask whether it needs to exist. Sometimes the most productive thing you can do is eliminate work that should not be happening rather than finding more efficient ways to do it.