Zero-Handoff

Never wait for anyone.

How would your work change?

Your best developer and your sharpest designer, ready the moment you need them. No scheduling, no handoff, no next sprint. You can build this today: you put an expert's judgment into a digital twin and work through it. One conductor, a team of expert twins.

A conductor on the left, an amber glow behind, raises a baton and sends orange cue lines to a row of green digital twins, each drawn with a dashed outline, while the real person each twin is modelled on stands faintly behind it.
Read

01 - What's new

You can build a twin of yourself.

For the first time, you can put what you know - your judgment, your taste, the way you would handle a problem - into an agent that carries it. A digital twin of you.

It works while you sleep, while you are in another meeting, while you are off entirely. And it is not only you: any expert can be encoded the same way, so their judgment is ready the moment anyone needs it.

That is genuinely new. Work has always needed the person present; now a twin stands in for them, so being in the room is optional.

Stop picturing AI as a tool you operate. Picture a version of you that works when you do not.

One solid dark figure with progressively smaller, lighter copies of itself working in parallel, an amber glow behind the first.
One of you, then many. Your judgment, encoded once, runs in parallel while you are asleep, in another meeting, or gone.

02 - The shift

Then the calendars become the bottleneck.

Hand the building to twins and it stops being slow. And the instant building is fast, the thing that holds everything up becomes obvious: people's calendars.

When a build took days, waiting an afternoon for the right person did not stand out. When a build takes minutes, that wait is the whole delay. So you encode the experts as twins too - always on, no calendar to book - and the waiting goes away.

What is left to improve is the communication itself: the handoffs, the lining-up and sign-off, the meaning lost when it passes between people. With building cheap, that communication tax is the whole cost, so cutting it is the whole job. Fewer handoffs, less waiting, fewer rounds, better output. Lean, in its original sense.

The next step in an old line.

It is the same idea each era had: cut the waste between the work. Zero-Handoff just applies it the moment building stopped being the hard part.

Then

Lean

Removed waste from the factory floor.

Then

Agile

Removed handoff waste from software teams.

Now

Zero-Handoff

Removes handoff waste once work agents make the building itself cheap.

Agile's ceremonies existed to manage handoffs. Twins remove the handoffs, so the ceremonies become pure overhead.

The best person, on every project at once.

Growing a company used to spread talent thin. You could never staff your best architect on every team, so seats filled with whoever was free. Encode the best judgment once and it runs on every project at the same time - your best work no longer has to be saved for a few teams.

A stack of ready-to-ship cards on the left, a clock marking the wait in the middle, and a calendar grid on the right almost entirely full with only one open slot highlighted.
Building is instant, so the work piles up ready. What it waits on is an open slot on someone's calendar - that wait, not the build, is now the whole delay.

03 - Two roles

Work splits into two roles.

When expertise is always available, the company settles into two jobs, and only two.

Width

Conductor

Owns a process from start to finish and runs it by directing expert twins.

Depth

Expert

Owns a craft and encodes it into twins for the conductors to use.

The product builder is simply the conductor of software development - one example of a wider pattern. A conductor is one person working with the reach of a whole department, because their best people stand around them as always-on twins.

Two jobs are left: orchestrate a process, or encode a craft. Everything in between was just handoff.

dev ux sec data infra domain conductor (you) expert twins · always on · never busy
You at the centre as the conductor, your best people around you as twins. The same judgment is available to every conductor at once.

Kept honest by three tiers of review.

A one-person setup is only trustworthy if its decisions are checked - and they are, scaled to how far a mistake would travel.

Tier 1 · low

Decide

The twin makes the call and the builder ships. No review. Most work lives here.

Tier 2 · medium

Assume, checked after

The twin goes ahead on an assumption it is allowed to make. An expert reviews afterwards and fixes it as they go. Nothing waits.

Tier 3 · high

Check first

Too big to undo. A human looks for a few minutes before it goes ahead.

And the experts are not going anywhere. They staff those checks, maintain the twins, and take the hard problems that now come up more often. The repeatable part of their work becomes a twin; their own time moves up to the hard and the new. A promotion, not a layoff.

Those checks are the human half of the checking. The automatic half - tests and known-good cases for each twin, and a spec that rejects broken work - runs alongside it. Together they replace the error-catching the old handoffs used to do for free.

04 - The honest part

The tax shrinks. It does not vanish.

Twins remove the waiting, but they do not make communication free. Moving an idea from your head into a twin still loses a little. What is left of the tax piles up at two points, and the human handoffs that used to quietly catch mistakes are gone.

stake- holder the builder twin interface 1 interface 2 capture intent the spec - old handoff checkpoints, now removed -
human (original) twin (representation) where the tax concentrates
The two points where intent passes across. Lose as little as possible at both, and add cheap checking back in without slow human handoffs.

A twin is incomplete in two ways.

Scope

A subset

It captures only part of an expert's skill - platform architecture plus some process design, not everything they know. This is why humans are still needed for the parts it doesn't cover.

Time

A snapshot

It is frozen at a moment and can go stale. This is why twins need looking after - and, later, it is exactly what keeps the expert valuable over time.

05 - The model

What the builder actually does.

The builder's one job that can't be handed off is holding the intent and fitting the twins' work into one whole.

It is not being an expert in everything. The builder owns a complete product, and what stays theirs is holding the intent and fitting the twins' work into one whole, plus making the calls the twins are not allowed to make. The twins work in separate crafts and rarely clash, so this is putting pieces together, not refereeing. A conductor, not someone who knows it all.

The engine underneath is a flip: specialists stop doing the work by hand and instead encode their judgment into twins that run on their own. The builder calls on them when needed. Under both sits the platform itself - a paved road of guardrails and ready-made paths that makes the default way to build the safe, fast one.

Someone builds and runs that platform: the AI architect. They design the twins, set the limits each twin works inside, fix the rules they all follow, and keep it all up to date. The builder builds; the AI architect makes building possible.

The conductor isn't the expert on anything. The conductor is the only one holding the whole intent.

How a twin is actually made

An AI model is easy to shape: tell it to answer in a made-up dialect or a casual tone and it does. That is the canvas. By giving it your decisions, judgment, and background knowledge through instructions, prompts, and context, you press a small slice of your expertise onto it. That slice is the twin. Encoding isn't fancy - it is just writing down the limited judgment a specialist already carries.

The supporting cast

The experts from section two, seen up close. Their job changes shape rather than disappearing. Each expert keeps their own twin current, maintains the limits it works inside, and still handles the rare cases it cannot: the new, the creative, the things you cannot write down. Their twins form one shared layer that supports every conductor at once - support teams, turned into twins.

Product types

What one builder owns

SaaS service or API · consumer app · internal tool · data product · platform component · regulated backbone · integration and middleware. One builder owns one complete product; the domain sets the core size.

The experts behind the twins

Who supplies the know-how

UX · security · data modeling · platform architecture · process design · compliance · performance and reliability · accessibility · domain SME. Each supplies a twin the builder can consult when needed.

Examples, not a fixed list. Which twins a product actually needs is decided field by field.

06 - The authority model

What is law, and what is yours.

A small core of rules is fixed law; everything else is a default the builder can override.

You have seen that bigger decisions get more checking. The deeper question is which decisions are even the builder's to make. A small core of unbreakable rules is fixed law, with no appeal. Everything outside it is a default the builder can override with one logged reason. That imbalance is the whole point.

And the line between the two is not fixed once and for all. Every logged override is evidence, and the line moves with it, so what counts as law versus preference is set by what actually happens, not decided in a meeting.

The further a mistake could spread, the more friction it should take: a few rules are law, everything else is yours to override in one logged click.

hard core Invariants · no appeal security, data/regulatory, cross-product Soft shell · defaults Builder sovereign override = one click + a logged reason friction increases inward →
hard core (invariant) soft shell (default)
The override log is what feeds the loop: it moves the line between law and preference from real use, instead of fixing it once. The core can shrink, not only grow, or the builder ends up caged.
blast radius → friction → soft shell 1 click + reason core no appeal
Things you can't undo - public API contracts, data migrations, promises to outsiders - pull a decision toward the core, where you plan carefully up front instead of trying things.

Decision spaces keep the human edge without the wait.

There is a trap hiding in "encode the common, stay human for the rare": every time you fall back to a human, the waiting quietly comes back. So the most valuable thing a twin does is not making every call - it is setting the boundary the builder can decide freely inside.

The expert hands the decision over in advance. The builder acts without waiting, and the twin can cover less. A partial twin still gets you most of the way.

07 - Working method

Features over stories. Capture cheaply, promote lazily.

The user story was a tool for handoffs; remove the handoff and it loses its job.

So drop to "I want this to do X" - and always keep the why, because the reason is what the agent needs to fill the gaps well. A feature with no why is the one the agent guesses wrong.

Keep work flowing, and limit how much is in progress at once, so you are not tempted to overbuild everything now that building is cheap. The pass/fail checks move from before to after: with the loop down to minutes, they take shape as you go, then get written down.

A wrong rule promoted too early costs more than all the duplication it was meant to remove.

The rule of three

Every builder decision could become a general rule, but do not sort each call on the spot - that brings back the friction you removed. Capture cheaply, promote lazily: turn a decision into a shared rule only when it comes up again and again across builders and products, not the first time. A wrong rule made too early costs more than the duplication it removes. Allow rules to be removed too - a rule builders keep overriding is a wrong rule.

08 - Bigger than product

The same shape, every process.

The two roles you met early were not specific to software.

The product builder is just one conductor, the one for the development process. Support, marketing, operations, finance, legal: every process can have its conductor directing expert twins, backed by the experts who keep those twins sharp. The same shape, repeated across the whole company. Not a better way to run a product team; a new way to organize knowledge work.

ProductSupportMarketingOperations leadership · mandate + backing
One conductor per process, each directing expert twins, all standing on a foundation of leadership backing. Dark is the conductor; green is a twin.

Companies win by backing their conductors.

A few sections ago this was a requirement. It is also the whole point. The companies that win are the ones that back their conductors with strong twins, real expertise, and a clear go-ahead, and then let them decide.

Leadership's job stops being to direct the work and becomes to clear the way for it. The moment managers climb back into the work, the tax they were trying to remove grows straight back.

Every process gets its conductor. Leadership's new job is to back them and then get out of the way.

Honest limit: this holds wherever judgment can be encoded and a process can be owned from start to finish. Work that is deeply unspoken, physical, or all about human relationships resists it. The reach is wide, not total.

Strip it all back and the whole model is one move: back your conductors and clear everything out of their way. Give them real twins, real authority, and the room to decide - the companies that do are the ones that win.