Picture an owner with a productivity book on her shelf that taught her EAD in 2018. She has reread the chapter twice in the last year. She has also read three things about AI as a personal assistant, watched a colleague rave about it on a Zoom, and bought a ChatGPT Plus seat that she uses maybe once a fortnight for a tricky email. The two reading lists have not met yet. The framework she trusts and the tool everyone is talking about are sitting in different parts of her head.
That gap is worth closing. The framework still holds. What has changed is which of the four nodes AI can credibly occupy, and that has caught a lot of careful operators on the back foot.
What is the EAD-Do framework?
EAD-Do is a four-step way of ordering any task or work pattern by impact: Eliminate it, Automate it, Delegate it, or Do it yourself. The order matters. You ask the questions in sequence, and the lower-impact answer is only correct if the higher-impact ones have all been ruled out. The first three buy you time at multiples; the last spends time at parity.
The lineage runs through Rory Vaden’s Focus Funnel, articulated in his 2015 TED talk “How to Multiply Your Time” and his book Procrastinate on Purpose: 5 Permissions to Multiply Your Time. Vaden’s original five permissions were Eliminate, Automate, Delegate, Procrastinate and Concentrate. The EAD-Do recast collapses the last two into a single Do node, on the basis that procrastinating-on-purpose and concentrating are both choices about when and how to do the work yourself, not separate categories.
The recast for 2026 changes one thing more substantial than the merge. AI is now a credible occupant of three of the four nodes, including the one nobody used to put it in. That is the move worth working through.
Where does AI sit in the Automate node?
Automate is the obvious AI quadrant and the one people picture first. This is where workflow software, integrations, and AI features inside existing tools remove repeatable steps the team currently does by hand. Invoice routing, calendar scheduling, meeting summarisation, document classification, customer-service triage, lead scoring. The shape is rule-bound work, predictable inputs, predictable outputs, no judgement call required at the centre.
The leaders are visible. McKinsey’s 2023 productivity-frontier work put the economic potential of generative AI at $2.6 to $4.4 trillion annually across enterprise functions, and the lion’s share lives in this quadrant. For an owner-operator the practical question is which two or three repeatable processes to give to a workflow tool first, what the setup time is, and what the unit-economics look like over a quarter. None of that is conceptually new. AI just dropped the floor on what the term “automatable” means.
The trap is sitting in plain sight. Automating something that should have been eliminated produces a faster version of work that did not need to exist, and the time saving disappears into more of the same. Worth running the first question first.
Where does AI sit in the Delegate node?
Delegate has changed more than any other quadrant. The classical move is to hand a task to a person, carrying hire cost, training cost, and management cost. The new route is to hand the task to AI, briefed the way you would brief a junior. Drafts, research summaries, formatting, comparison tables, first-pass analysis. AI now does it in the time it takes to write a clear brief.
This is what the delegation maths now looks like. The cost-per-output curve has flattened, the unit time has compressed, and the briefing skill has become the bottleneck. Wade Foster, Zapier’s CEO, has documented his personal AI workflow on Lenny’s Newsletter as exactly this: AI as a delegate that needs a clear brief, not a magic wand. The pattern is consistent across operators who have published their stack.
What it does not change is the management muscle. AI delegates need the same thing a person needs at the start: clear scope, clear definition of done, and review of the output. The founder who never learned to delegate to people does not get to skip that lesson by adding AI; they hit the same wall, just at a different scale. Worth treating your AI questions as the same questions you would ask a contractor.
Where does AI sit in the Do node?
Do is the surprise. The classical reading is that this is the founder’s own concentrated work, the things only she can do. The recast says AI sits inside that work too, as a thinking partner alongside the founder rather than a substitute for her. Ethan Mollick’s Co-Intelligence and his “cyborg writing assistant” piece name the pattern directly. AI compounds the founder’s thinking, it does not replace it.
The public examples have stacked up. Reid Hoffman co-wrote Impromptu with GPT-4 and made the thinking-partner relationship visible from inside the prose. Aaron Levie at Box has talked publicly about AI as collaborator on TechCrunch and on the Latent Space podcast. Andrej Karpathy has tweeted about AI as cognitive prosthesis. None of these people are using AI to write their work for them; they are using it to stress-test ideas, run pre-mortems in the Gary Klein sense, generate counter-arguments, and surface what they have not thought of yet.
The discipline matters. The Mata v. Avianca case from 2023, where lawyers cited fake AI-generated case law in court, is the standing reminder that AI in the Do node still requires human verification at the output layer. AI as thinking partner is high-value. AI as unverified output is a fast route to embarrassment.
Why does Eliminate stay human?
Eliminate is the only node AI does not occupy. The decision about whether a piece of work should exist sits in the founder’s strategic context, and AI cannot infer that from outside the business. The quarterly client report nobody reads, the meeting cycle that lost its purpose, the product line that stopped paying for itself, those are judgement calls anchored in what the business is for.
What AI does change is the visibility of the invisible work. Many owners cannot tell you, accurately, where their week went. They can tell you the headline projects, but not the thirty-five small recurring tasks underneath. AI now does calendar audits, decision-log mining, and time-tracking analysis well enough to make the underlying pattern visible. With the pattern visible, the owner can make the elimination call on real information rather than on memory and feel. That changes the quality of the decision, not who makes it.
Skip this node and the rest of the framework misfires. Automate without eliminating and you produce faster waste. Delegate without eliminating and you train AI on work that was not worth doing. Do without eliminating and you spend founder hours on tasks that should have been killed last quarter. Run the Eliminate question first, every time, even when it slows you down. The whole framework is an impact hierarchy, and the top of the hierarchy is still a human call.
If you want to talk through where each of the four quadrants lands in your own week, book a conversation.
