Delegate the high-frequency, low-judgment tasks first: research, follow-ups, reporting, and triage. Save the relationship moments and high-stakes decisions for last, or never. The order matters, because starting in the wrong corner is how founders get burned and conclude AI "does not work."
Key takeaways
- Plot every task on two axes: how often it happens, and how much judgment it needs.
- Start in the high-frequency, low-judgment corner for the best return at the lowest risk.
- Delegate frequent-but-judgmental tasks with tight human approval on every send.
- Keep rare, high-stakes, relationship-led work human, or have an agent only prepare you for it.
- Starting with the "interesting" task instead of the valuable one is the classic mistake.
There is a simple way to see it. Plot every task on two axes and the answer jumps out. This is the prioritisation companion to how to get 10 hours a week back and the broader pillar guide.
The delegation 2x2
Two questions decide everything. How often does this task happen? And how much judgment does it need?
| Low judgment | High judgment | |
|---|---|---|
| High frequency | Delegate first (research, triage, reporting, chasing) | Delegate with tight approval (proposal follow-ups) |
| Low frequency | Delegate when convenient (occasional admin) | Keep human (big negotiations, key hires) |
The top-left box is your starting line. High-frequency, low-judgment tasks are frequent enough that automating them saves real hours, and simple enough that the agent rarely gets it wrong. Best return, lowest risk. The other three boxes are not "never," they are "later" or "carefully," and knowing which is which stops you from either over-automating or missing easy wins.
What lives in the "delegate first" box
The usual suspects, and they are usual for a reason. Prospect and market research. Inbox triage and drafted replies. Invoice chasing and follow-ups. Weekly reporting and KPI monitoring. CRM updates. Meeting notes turned into actions. Every one is something you do constantly and could describe to a new starter in a paragraph. If you are staring at your week wondering where to begin, begin here, because these tasks combine the two things that make delegation safe and worthwhile: they happen often, and a mistake caught in a draft costs nothing.
What to delegate carefully
The high-frequency but high-judgment tasks, like following up a stalled proposal, are worth delegating, but with a human approving every send. The agent does the reading and drafting; you make the call. You get the speed without gambling the relationship. This is human in the loop applied deliberately: the more judgment or reputation is at stake, the tighter the approval, so an agent can help with sensitive work as long as you stay firmly at the decision point.
What to keep human
The rare, high-stakes, relationship-led work. The negotiation that makes or breaks the year. The key hire. The delicate client conversation. An agent can prepare you for these, brief you, draft your notes, but it should not run them. Relationships are the one thing you do not outsource to software. Keeping these human is not a limitation of the technology; it is a deliberate choice about where your own judgment is the entire value, and a good AI setup protects that rather than eroding it.
Why this order beats "automate the exciting bit"
Founders love to start with the clever, novel task because it is interesting. It is also usually low-frequency and high-judgment, the worst corner to begin in. You spend weeks on something that saves an hour a month and occasionally embarrasses you. Start boring, bank hours, build trust, then get ambitious. This is exactly the order a good audit walks you through, because it is the order that actually works, and it is the same discipline behind what to automate first.
A tangible example: imagine Ravi choosing where to start
Imagine Ravi, a founder itching to build an impressive AI sales-strategy assistant. He plots his tasks on the 2x2 first and sees that strategy sits in the rare, high-judgment corner, while inbox triage, invoice chasing, and weekly reporting sit squarely in the delegate-first box. He resists the shiny option and starts with the boring three. It is realistic to picture Ravi reclaiming a day a week within a month and building real trust in his agents, whereas the strategy bot would have absorbed weeks for a marginal, occasional gain. Later, with hours banked and confidence high, he adds the ambitious stuff. Starting boring is what made the exciting possible.
Turning the order into a plan
Once you have plotted your tasks, the plan writes itself: number the delegate-first box by hours saved and work down the list, one agent at a time, proving each over a fortnight before adding the next. If you would rather have that plan built with you, a free audit does exactly this plotting and sequencing, then builds the agents in order. Start on the home page.
Reporting: a perfect first delegation
If you want a single task to prove the priority order, pick reporting, because it sits right in the delegate-first corner and the payoff is obvious. A weekly or monthly report is frequent, it follows a describable shape, and getting it slightly wrong in a draft costs nothing because you review it before anyone sees it. Hand it to an agent and it pulls the numbers from your tools, assembles them into the format you use, and writes a short plain-language summary of what moved, ready when you sit down. Founders are often surprised how much time this quietly reclaims, because report-building is the kind of work that does not feel huge in any single instance but adds up to hours across a month. It is a low-risk, high-relief first delegation that builds your confidence for the rest.
Common delegation mistakes
Three mistakes trip founders up when they start delegating to AI. The first is starting in the wrong corner, reaching for the impressive, rare, high-judgment task instead of the boring, frequent, safe one. The second is delegating a judgment task without keeping a human on the send, so an agent fires off something tone-deaf and the founder loses trust in the whole idea. The third is trying to delegate everything at once, a big-bang rollout that overwhelms both the founder and the setup. All three are avoided by the same discipline: start in the top-left box, keep approval on anything sensitive, and add one agent at a time. Get that right and delegation feels safe and cumulative rather than risky and chaotic.
A quick self-check before you start
Before you delegate a task, run it through three quick questions. Does it happen often enough that automating it saves real time? Could you describe how it is done to a new starter in a paragraph? And if the agent got it wrong, would you catch it before any harm was done? Three yeses and it belongs in your first wave. If any answer is no, the task either waits or stays human. This thirty-second check keeps you honest and stops you from either automating something you should not or hesitating over something that is obviously safe. Run every candidate task through it and your delegation order almost designs itself.
