Sixteen years ago, during an internal training programme, I put
three problems in front of a large group of colleagues and asked each of them
to pick one.
The first was long and tedious. It needed a lot of brute-force
computation, but the method was clear from the start. The second was a logic
puzzle. The answer was measurable and there was obviously a method to reach it,
though you had to find that method yourself. The third had no clear method and
no single right answer.
Most people picked the first. A few picked the second. Only a
handful picked the third.
I wrote about that experiment on my blog in 2010, in a two-part post
on scaling up. My argument then was about managing uncertainty inside
organisations. I still hold that argument. But the years since, and especially
the last three, have convinced me that the more important story is about that
handful of people who reached for Problem 3, and about what made them
different.
We are trained
to choose Problem 1
The choice was not surprising. Most of us spend our lives buying
certainty. We accept the uncertainties we cannot control, such as health,
weather and fate, with some equanimity. Where we do have some control, we work
hard to remove uncertainty.
We go to college partly so that a degree can vouch for us. We join a
company instead of starting one because someone else has already decided the
business, the product, the technology and the process, and our job is to
execute. A government career reduces uncertainty further, because in many
positions the system asks for compliance with process far more than it asks for
results.
There is nothing wrong with this. It is how societies function. But
it means that by the time most people reach a workplace, they have spent twenty
years being rewarded for choosing Problem 1. Exams have answer keys. Appraisals
have rating scales. The habit of reaching for the problem with a clear method
is not a personal weakness. It is what our education and our institutions have
trained into us.
The machine for the many
In 2010 my conclusion was practical. If most people are
uncomfortable deciding under uncertainty, then an organisation that wants to
scale must reduce the uncertainty they face.
The small founding team of any venture is usually made up of
restless, high-energy people who are at ease with ambiguity. They cannot run
the operation at scale, and they rarely want to. Scale depends on large numbers
of people who do their best work when the task is clearly defined.
So you map organisational goals down to team deliverables and then
to individual deliverables. You write standard operating procedures and
business rules for as many cases as you can. You build in exception reporting
and escalation paths for the cases nobody anticipated. Without these, people
behave like software with no error handling. They hang.
Chris Argyris made a similar point in Harvard Business Review,
observing that reengineering worked best when jobs were rigorously specified
rather than left for individuals to define. I also warned about the opposite
failure: teams that follow the process so faithfully that they forget the
result it was meant to produce.
All of that still stands. Systems, processes and SOPs carry the many
through what is known. Every institution that has scaled, from a bank branch
network to a national digital platform, has done it this way.
What
changed: the machines took Problem 1
In 2010 I assumed that Problem 1 work would always need people. That
assumption no longer holds.
Long computation with a known method is exactly what software has
always done well. What is new is that AI systems now handle a growing share of
Problem 2 as well. Give a capable model a well-posed puzzle, a contract to
review against a checklist, a dataset to reconcile or code to write against a
clear specification, and it will often find the method faster than most of us.
A well-written SOP was always a kind of program for people.
Increasingly it is becoming a program for machines. The very work that
organisations spent decades making predictable is the work that is easiest to
automate, because it is predictable.
This is the uncomfortable part. The people who chose Problem 1 for
their careers did nothing wrong, but the ground under that choice is shifting.
And the scarce resource in any organisation is now the person who can face
Problem 3, a situation with no method and no single right answer, and still
move forward.
AI is useful there too, as a research assistant, a sparring partner
and a fast drafter. But it does not decide which problem is worth solving, or
own the consequences of a judgement made with incomplete information. That
remains a human job.
What Problem 3
asks of a person
What carries a few people into Problem 3 is something no SOP can
specify. In my experience it has three parts.
The first is critical thinking: the ability to question how the
problem has been framed. Problem 3 situations rarely arrive labelled. They show
up as a stalled project, a market nobody serves, or a regulation that works on
paper and fails on the ground. The first act of problem solving is often to ask
whether we are solving the right problem at all.
The second is the discipline of problem solving itself. Because
there is no algorithm, you have to build one: break the mess into parts, test
assumptions cheaply, accept partial answers and improve them. This is less
glamorous than it sounds. Most of it is patience.
The third is an inventive mind, one that is willing to try an
approach that has not been tried and is comfortable being wrong in public along
the way. Most organisations punish visible failure far more than invisible
inaction. People learn quickly that choosing Problem 1 is the safer career
move.
None of these is rare in the sense of being a gift. Each can be
trained and strengthened. What is rare is an environment that rewards them.
The
advantage of having been elsewhere
There is a fourth quality that I have come to value more than any of
these: exposure to diverse domains, and the ability to distil a lesson from one
field and apply it to another.
Problem 3 in one domain is often Problem 1 in another. Someone has
already solved something structurally similar. The difficulty is that they did
it in a different industry, using different vocabulary, and nobody in your room
has ever been there.
Taiichi Ohno built the pull system at Toyota by borrowing the logic
of the American supermarket, which restocks shelves only as customers empty
them. According
to one history,
he was applying the idea in his machine shop by 1948, years before he actually
visited a US supermarket. Closer home, Dr G. Venkataswamy built Aravind Eye
Hospital’s high-volume model of cataract surgery after studying how
McDonald’s achieved scale.
A fast-food chain taught an eye hospital how to restore sight to millions.
India’s digital public infrastructure has followed the same pattern.
The idea that commerce could be opened up the way UPI opened up payments, with
a shared protocol and many interoperable players instead of a few closed
platforms, came from carrying a lesson across domains. It did not come from
inside the industry it eventually changed.
David Epstein makes the broader case in his book Range: in
complex, unpredictable fields, generalists who have sampled widely often
outperform early specialists. That suits a world where AI is becoming the
ultimate specialist.
The real job of the few
This is where my two arguments, sixteen years apart, meet.
It is tempting to read all this as a story of heroes and followers,
with brilliant minds on one side and process-bound masses on the other. That
reading is wrong, and it is dangerous.
The people who pick Problem 3 do not matter because they avoid
process. They matter because they create it. Their real contribution is to turn
Problem 3 into Problem 2, by finding the shape of the problem, and then into
Problem 1, by writing down the method so that thousands of others can follow it
without needing to be heroes.
Every SOP that works was once somebody’s Problem 3. Somebody faced
the first failed transaction, the first angry customer or the first case the
rules never anticipated, and worked out what to do. The SOP is the record of
that thinking, written so that nobody has to repeat it.
Seen this way, the few and the many are not in competition. They are
two halves of one system. An organisation with only Problem 3 people never
scales. An organisation with only Problem 1 people stops working the moment the
world changes, and in 2026 the world changes every quarter.
The turbocharger
Skills explain how someone solves Problem 3. They do not explain why
anyone would choose it in the first place, or stay with it once it turns
difficult.
Problem 3 is lonely for a long time. The early evidence is thin, the
critics are confident, and every rational signal says retreat to safer ground.
Many capable people with all the right skills turn back at exactly this point.
The ones who keep going usually have three things that no training programme
can install: commitment to the work, conviction that the direction is right,
and a quiet faith that the thing is worth doing even if it fails.
I use the word quiet deliberately. The faith I mean is not the loud
certainty of the pitch deck. It is closer to the steady belief that keeps
someone working on a hard problem after the meeting ends and the audience has
gone home.
When I wrote in 2010, I ended with a line from Bruce Barton, that
nothing great has been achieved except by those who dared to believe something
inside them was superior to circumstances. I would put it differently now.
Skills are the engine. Commitment, conviction and that quiet faith are the
turbocharger. An engine without one still runs, but it will not climb the
steepest roads.
What
institutions should do differently
If Problem 3 thinking is now the scarce resource, most organisations
are set up to waste it. Five changes would help.
1.
Hire for range, not only for
depth. A candidate who has worked across sectors,
or moved between government, markets and technology, is often screened out as
unfocused. In a Problem 3 world, that breadth is the asset.
2.
Give people Problem 3 work
early, and protect them when it fails. Comfort with
uncertainty is built through practice. If the first honest failure ends a
career, nobody will try a second time.
3.
Measure people on results,
not just on compliance with process. An SOP is a
tool for producing an outcome. People who improvise sensibly within agreed
boundaries to reach that outcome should be rewarded, not written up.
4.
Promote for judgement under
uncertainty. Many organisations take their best
technical expert from a low-ambiguity role and place them where ambiguity is
the whole job. They often struggle, and the whole team suffers with them.
5.
Let AI handle Problem 1, and
redeploy the time. The productivity gained from
automation should not simply become headcount savings. Some of it should buy
capacity for people to work on problems that do not yet have a method.
The same applies to our education system. As long as our exams
reward the fastest route to a known answer, we will keep producing graduates
who instinctively pick Problem 1, just as machines are making that choice
redundant.
Which problem would you pick?
When I first wrote about the
experiment, a reader asked in the comments whether I would share the three
problems. I never did. Perhaps that was the right instinct, because the
specific problems never mattered. The choice did.
If I ran the same exercise today, I
suspect the result would look much the same. Most people would still reach for
Problem 1. The difference is that in 2010 that was a safe choice. In 2026 it
means competing with a machine on the machine’s own ground.
Systems will carry the many through what
is known, and they should. Institutions that scale have to be built that way.
But the future of those institutions, and of the people in them, will be
decided by the few who reach past the clear method for the problem that has
none. They will bring with them sharp thinking, lessons borrowed from other
fields, and enough quiet faith to keep going.
The good news is that the few do not have
to stay few. The question is whether we will build organisations and schools
that make room for them.
“You can’t connect the dots looking
forward; you can only connect them looking backwards. So you have to trust that
the dots will somehow connect in your future.”
Steve Jobs, Stanford commencement
address, 2005
This essay builds on two posts from
May 2010: Scaling
up: The Art of the Impossible, Part I and Part
II.
