Which Tasks Should Be Delegated to AI — and Which Never?

What Is on the Agenda?

The delegation question arrived at the same table through four doors.

Potential research drew the ceiling

McKinsey’s activity-level arithmetic finds that work absorbing 60-70 percent of employees’ time is technically automatable — with the source’s own caveat: the ratio belongs to activities, never whole roles. The ceiling is high; the decision still lives in single tasks.

Behaviour measurement showed the floor

Real-usage measurement finds delegation concentrating in defined, narrow tasks: 77 percent delegation patterns in enterprise embedding against a near-even split on the personal surface. Potential and behaviour reach the same verdict by different roads: delegation is set by the work’s drawing, not the machine.

The capability ruler started moving

The task length machines finish alone doubles on a regular clock. Yesterday’s “undelegatable” drifts into today’s “together” and tomorrow’s “delegate” — a business carrying a fixed guide errs in both directions at once.

Accidents sharpened the question

The early delegation wave’s accidents — automatic replies gone wrong, ownerless commitments, unrecallable actions — moved “what shouldn’t we delegate” from theory to case file. Those files are the most honest raw material of this set’s never list.

Why Is the Question Being Asked Now?

The question is old; four pressures made it undeferrable.

Tools stopped carrying their mode

Old automation wore its nature openly: the conveyor delegated, the spreadsheet assisted. Today’s tools open onto every kind of work from one box; the tier is set by a user decision, and when the decision turned invisible it started going unmade — an unmade decision shows its worst face on the busiest day.

Waste piled up in both directions

Work that should have been delegated stayed on human shoulders and cost efficiency; work that should never have been delegated went out and cost reputation. Both wastes billed the tool, yet both were tier errors. As the files thickened, the question’s address moved from the tool catalogue to the management table.

Teams demand clarity

A tierless “AI is coming” grows exactly one thing: uncertainty. Written down — these tasks delegate, in these the machine sits beside you, these stay human in every case — resistance turns to participation. Clarity is this period’s cheapest motivation instrument.

Clients ask about the border

In a market of cheap production, the client’s question became: who stands behind this? The business that can state, in writing, “on these decisions the last word is always human” collects a quiet trust premium — the border became a declaration with a market price.

What Is Wrong?

Four maxims circulate, all carrying the decision to the wrong unit.

“These professions delegate, those are safe”

A profession is a weave of dozens of tasks; a weave takes no single stamp. Every meeting held at profession level polarises — every meeting lowered to activity level produces a list. The unit of decision is the task; this set’s entire ruler is built on that unit.

“If the tool is good enough, delegate”

The tool’s skill is not even one full axis of the ruler. Unverifiable output, irreversible consequence and judgment load block delegation in the best tool too. A decision starting from the tool hunts work for the tool; one starting from the task picks tools for the work.

“Human oversight fixes everything”

An approval pressed unread is décor, not oversight. “Do together” is not a good intention but a construction: a zoned line, an approval reduced to one question, a logged rejection. Without flow design, human oversight turns to stamping in the first heavy week.

“The never list holds until technology catches up”

Never and not-yet are different lists: not-yet is a capability list, alive through tests and dates; never is a responsibility list, independent of capability. Even on the day the machine signs perfectly, the signature does not delegate — the border’s rationale has its own piece; its principle is one sentence: delegation carries execution, never accountability.

The Real Mechanism

The tier arrangement has four parts; none substitutes for another.

The ruler: three axes

Every task reads on three axes. Length-scope: how many hours of a competent person? Verifiability: is checking the output distinctly cheaper than producing it? Price-judgment: can an error be undone, and does the task carry signature, commitment or a value choice? The three axes in practical form are the five-criterion test — ten minutes per task, result one of three tiers.

The tiers: three contracts

The delegate tier runs on an audit contract: sample, fault log, threshold-alarm, stopping authority. The together tier on a flow contract: preparation in the machine, decision in the human, rejection on record. The never tier on a responsibility declaration: the last step human in every case. A tier without its contract is not a tier but an intention — and intention evaporates on the first exception day.

The gates: two-way transfer

Movement between tiers is ruled: a together-task whose rejection rate stays under threshold promotes to delegate; a delegated task whose audit cost approaches its gain steps down to the flow; the not-yet list re-scores on a calendar. Ruled gates turn the guide from a one-time verdict into a living arrangement — the mode framework builds the gates’ logic in full.

The address: a plan for the freed hours

The mechanism’s most-skipped part: where delegation’s recovered time flows. Unaddressed time drains into scatter and the delegation reports as “slack” rather than “gain”. The time’s address is written with the delegation decision — which judgment work, which client contact, which improvement. That is distillation’s precondition: swollen time to the machine, dense value to the human.

Who Is Affected, and How?

The same arrangement produces four different first moves across four profiles.

The knowledge-heavy service business

This wave’s first address: desks that write, read and answer. The task inventory is rich, the delegation candidates plentiful — and the commitment load high. The first job is slicing: preparation layers to delegation and flow, binding sentences to the never lines. Sliced correctly, the service business gains capacity without spending quality.

The business whose processes live in heads

Where tasks are unwritten there is nothing to test, and the delegation debate spins in air. The precondition is process writing — an investment that repays independently of AI: handover, training and audit all feed from the same text. In an unwritten business, what happens is not delegation but abandonment.

The small, single-decider structure

With signature and execution in one hand, the border blurs fastest here: the one person delegating everything begins, unnoticed, to delegate their commitments too. The three-line never list — price, commitment, irreversible send — is the small structure’s whole arrangement, and the one that needs it most.

The consultant and the agency

The arrangement’s second licence is consulting: the inventory-test-three-lists order works as a diagnostic frame at the client table. The consultant who talks catalogues sells products; the one who talks rulers becomes a decision partner — the difference reads in the relationship’s lifespan.

Decision Order

The arrangement builds in four steps; each grows dearer out of order.

One: inventory and test

The desks’ weekly tasks reduce to one-line definitions and pass through the five-criterion test; every verdict takes a date. This half-day round replaces months of “strategy” meetings — the strategy is nothing beyond the three lists’ sum.

Two: sign the never list

The four rationale columns — signature-commitment, value judgment, irreversible consequence, accountability — get written in the business’s own words; tasks touching a column go to the never lines, the rest gets sliced. The list takes force under management’s signature and is said out loud: the border is drawn first so that everything else delegates with a clear conscience.

Three: build each tier’s contract

The first delegate-task gets its audit page, the first together-task its flow page — each one sheet, each ready in its own tip. Two live examples teach faster than concept training; the organisation grasps the difference from comparison.

Four: connect the gates and the calendar

Promotion-demotion thresholds get written, the not-yet list takes its six-month re-scoring date, and the freed hours’ address joins the decision. With this step the guide stops being a document and becomes an operating arrangement — and when the next tool wave arrives, the business discusses its own ruler, not the tool.

Where to Start?

The first month holds four finishable jobs.

Week one: the photograph and the if-only list

A mode map of actual usage — who runs what, under what arrangement — plus the past year’s “human approval would have prevented this” events on one list. Two documents, the guide’s honest raw material: one shows the present state, the other the border’s necessity.

Week two: the test round

The busiest desk’s twenty tasks go to the test; three lists are born. The round’s by-product arrives for the first time in most businesses: the tasks’ written definitions — an output as valuable as the test itself.

Week three: two example builds

One delegated task goes live with its audit page, one together-task with its flow page. The aim is not just two tasks’ gain but the visible existence of “this is how the arrangement works” — an example travels faster than an instruction.

Week four: first reading and the announcement

The two builds’ first numbers get read, the never list goes to the team in one paragraph, the not-yet list takes its date. At week four’s end there is not a guide but a working arrangement’s first round — the guide is that round’s minutes.

What Not to Do?

Four reflexes break the arrangement before it forms.

Starting from the tool

“Buy the tool first, find the work later” bends the inventory to the tool and surrenders the ruler to the demo. The tool is the decision order’s last step — where it is also the easiest: picking a tool for work whose task, measure and contract are fixed is a technical comparison.

Ruling wholesale

“We delegate everything” and “everything here passes through a human” are the same error’s two ends: an ideological verdict issued without looking at the task. The tier is a measurement result, not a creed — and measurement surprises both camps with their own lists.

Delegating without a contract

Unaudited delegation looks flawless in its first weeks; the decay is silent and the invoice arrives in bulk. Delegation’s first day is the audit’s first day — no working brakes, no right to accelerate.

Setting a number target

“Ten tasks delegated this quarter” manufactures pressure to bend thresholds and removes the test’s fuse. Delegation counts are never targeted; they occur when conditions are met. The target goes on the arrangement’s construction — how many tasks were tested, not how many passed.

What to Watch?

Four gauges read the arrangement’s health from outside.

Test coverage

What share of tasks has been struck against the ruler and bound to a dated verdict? That single ratio summarises the business’s delegation maturity — coverage is discussed before tier distribution.

Tier movement

Per period: how many tasks promoted, how many stepped down, how fast is the not-yet list melting? A motionless map signals either a stagnant business or an unmaintained arrangement; a healthy one moves a few tasks every round.

Audit and rejection data

The fault logs’ class distribution and the flows’ rejection rates read together: falling rejection says maturing, rising material faults say decaying, a single critical fault says stop. This panel is the delegation’s conscience.

The freed hours’ realisation

The address was written; did the flow happen? The plan-versus-actual of delegation’s recovered hours is the arrangement’s final report card — if time is not turning into value, the delegation won only in the story, not in the books.

How Does This Period End?

The ending reads as three separations.

The era of wholesale verdicts closes

As “AI will end these professions” headlines age, the task-level minority’s language becomes the standard. Early movers finish building their arrangements before the argument ends.

The border declaration commercialises

“The last word is human” becomes a trust clause written into contracts and proposals. The business without a border competes on price; the one with a border competes on trust.

The gap compounds

Test habits, log data and gate records accumulate period over period: the builder speeds up each round, the non-builder restarts at every tool wave. The difference is compound interest, not addition — and compounding cannot be repaid, only pre-empted.

The question relocates

“Which tasks delegate” becomes, in the arrangement-building business, “which task is next this quarter”. The question’s shrinking is the sign the guide has done its job.

A Solid Digital Foundation

The tier arrangement stands on four stones, each outliving any tool.

A written task inventory

Tasks with written inputs, outputs and “what good looks like” — the shared base of the test, the contracts and every handover. Writing is delegation’s passport.

One record panel

Test sheets, fault and rejection logs, promotion-demotion decisions and exception records live in one place. Scattered records starve the ruled gates to death.

A named responsibility map

Opposite every never line, every audit threshold and every stopping authority stands a name. A nameless arrangement hunts for an address on the first crisis day — and that day, addresses are used, not hunted.

A periodic maintenance round

Every six months: the not-yet list re-scores, thresholds get read, the if-only list refreshes, time addresses compare against actuals. The round keeps the guide alive — and each round asks one double question: what did this arrangement protect us from last period, and what did it get us to in time? An arrangement with both answers defends its own budget.

Frequently Asked Questions