Preloader

Subtitle: The Eurostar spent billions to save 40 minutes. For 0.01% of that, it could have made the journey better. That is the whole problem of AI, jobs, and value in one story.

The 60-second reel that explains the next 10 years

A few days ago, a short reel of RorySutherland at TEDx Athens started circulating again on Instagram. Watch the reel here: rorysutherlandofficial89

He tells one story. That story is the best explanation I have seen for why we are stuck, and what System-Level Context Alignment Engineering, or SLCA, is for.

The Eurostar spent billions to cut the Paris to London journey by about 40 minutes. Engineering, tunnels, tracks, speed. A massive, logical, spreadsheet-approved achievement.

Sutherland asks a different question: What if we had spent just 0.01% of that money on Wi-Fi on the train?

The journey would not have got shorter. It would have got better. Passengers could work, watch, talk, relax. The time would feel useful, not wasted. He goes further, as a joke that is not a joke: Give everyone free luxury wine and supermodels to serve it, and you would still have billions left over, and people would prefer the slower train.

His point is not about trains. His point is about an asymmetry:

Mechanistic, engineering, logical ideas are accepted without proof. Psychological, creative, context ideas have to prove themselves with cost-benefit analysis, ROI studies, and feasibility reports before anyone will try them.

That asymmetry is now the central problem of the AI age.

Because AI is the ultimate mechanistic machine. It will make the train faster, cheaper, and more efficiently than we ever could. If we keep asking it only to make the old train faster, we will spend billions to save minutes, while missing the 0.01% shift that would create real value.

This is why I wrote in my previous article, Why There Is No AGI Finish Line To Wait For, And Why That Is Good News For Job Creation, that there is no finish line. We keep moving the finish line because we keep asking the wrong question: How much can a system handle?

The right questions are: What frame is it working in? Who chose that frame? What does the frame hide? What value is it trying to make visible?

The Eurostar reel is the same argument, in 60 seconds, with a train.

The two equations that explain the reel

Let Q be the question, C be the context, and V be the value that is recognised.

We live inside two relations, both valid, both needed:

1. Value = f(question, context) : V = f(Q,C)

Give the same question a different context, and the value you can read changes.

2. Context = f(question, value) : C = f(Q,V)

Start from the value you aspire to, then build the context where that value can appear.

Most of our education, our companies, and our AI benchmarks live only in the first equation. We are taught to enter someone else’s context and prove value inside it. A syllabus. An exam. A hiring process. A benchmark. A speed target.

Sutherland’s Eurostar is V = f(Q,C) in action.

Q is the same: How do we improve the Paris to London journey?

In C1, the engineering context: Value means time saved. So V1 = 40 minutes saved for billions spent.

In C2, the psychological context: Value means enjoyment and usefulness. So V2 = a journey that feels shorter, richer, and more productive, for 0.01% of the cost.

Same Q. Different C. Different V. Nothing about the train changed. What changed was what we decided was valuable.

And this is the second equation, C = f(Q,V), in action:

If you start with V = happy, productive passenger, you would never select C = dig a faster tunnel. You would select C = Wi-Fi, comfort, good coffee, quiet, space to work, a pleasant human experience.

The context is selected because of the value at stake. That is not a new idea. It is what good designers, teachers, founders, and parents do every day. What is new is that we never formalised it, and so we never taught it as a skill.

SLCA is that formalisation.

What is SLCA Engineering?

System-Level Context Alignment Engineering to the Value We Aspire To, or SLCAE, is the deliberate design of the frame an agent works in, so that what it already holds becomes recognisable and what it lacks becomes visible.

It applies to a person, a team, and a model in exactly the same way.

SLCA does not ask: How do we make the model smarter?

It asks: What context should this model, this person, this team be placed in, so that the value we aspire to can appear?

That is a system-level question, not a prompt-level trick. A prompt is one small part of C. A full context is a bundle:

  • The framing of the question
  • The tools allowed
  • The data shown
  • The audience
  • The metric for success
  • The time allowed
  • The examples given
  • The checker used
  • The role assigned

Change any one, and you have a new context. The combinations are effectively unbounded. That is why “we have run out of jobs” is the wrong diagnosis. We have run out of given contexts. We have not run out of contexts we can build.

The Eurostar engineers were excellent at V = f(Q,C). They optimised inside the given context perfectly. No one was paid to do C = f(Q,V). No one was paid to ask: Is speed the value we actually want?

In the AI age, that job is the most valuable job there is.

Why we cannot continue growing with the same old context

This is the need of the hour, and it is urgent for three reasons.

1. The old context has diminishing returns.

Billions for 40 minutes is the pattern of all mature systems. In AI, we see the same pattern now: Billions for a few more points on a benchmark, a little more speed, a little more fluency. The cost curve is steep. The value curve is flat.

We are optimising the wrong C.

A school that spends lakhs to make students 5% faster at an exam is the Eurostar. A school that spends a fraction to change the context, from “answer under stress” to “explain to a younger student,” will often discover a student who looked weak is actually deep, articulate, and strong. Same student. Same Q. Different C. Different V. I gave this example in the AGI article, and every teacher knows it is true.

2. AI is brilliant at V = f(Q,C) and blind at C = f(Q,V).

Give an AI a context, and it will produce value inside it faster than any human. Give it a question without a context, and it will assume the most common context from its training data. It will not ask: Is this the right frame? Is there a cheaper, more human frame that would create more value?

That is not a flaw to be patched with a bigger model. It is a structural limit. A model cannot know what value you aspire to unless you tell it. And you cannot tell it unless you know how to choose a context.

In my own small studies, reported in the AGI article, a supplied context beat a model-chosen context by 24 to 2 for one model and 16 to 1 for another on 100 questions. The model did not get smarter. The frame got better.

That is the Eurostar lesson for AI: The frame beats the engine.

3. The finish line keeps moving because the frame keeps changing.

In the AGI article, I argued that AGI has no finish line because any thinker has finite capacity against an unlimited space of things worth knowing. There is always a remainder: Unknown and Unrecognised.

Sutherland shows the same thing from the other side: Even when we “finish” the engineering problem, we discover the psychological problem was the real one. We finish the tunnel, and we realise we wanted Wi-Fi.

If we keep defining progress as “faster in the old context,” we will always feel behind. If we define progress as “better at shifting context to create value,” possibilities multiply.

The Ignorance Awareness Factor and the Recognition-Action Taxonomy in the train story

Two tools from my work make SLCA practical, not philosophical.

The Ignorance Awareness Matrix comes from two questions: Do you hold the knowledge? And do you recognise that it applies?

RecognisedUnrecognised
KnownKR: Known and Recognised. What you have and can use.KU: Known but Unrecognised. You hold it, but you do not see that it matters here.
UnknownUR: Unknown and Recognised. A gap you can see.UU: Unknown and Unrecognised. A gap you cannot see.
The Eurostar team was in KR for engineering. They knew how to build speed, and they knew it applied.

They were in KU for psychology. They held the knowledge that people value enjoyment, usefulness, and comfort. They just did not recognise that it applied to a train journey. The Ignorance Awareness Factor, IAF, is the practice of using behaviour, not self-report, to place a pattern in one of these boxes, then moving it on purpose.

KU is the cheapest place to start. The knowledge is already there. You do not need a new tunnel. You need a new frame.

The Recognition-Action Taxonomy, RAT, is the sequence:

  1. Recognise the context you are in
  2. Recognise what is valuable in that context
  3. Act to create that value

Most people, and most AI systems, skip recognition and jump to action. That is why they do the right task in the wrong context. They build a faster train when the passenger wanted a better journey.

SLCA Engineering is RAT made into a discipline: Do not act until you have named the C you are in, named the V you aspire to, and checked if a different C would produce that V cheaper, faster, or more humanly.

Every time you converse with an AI platform, it generates tokens. Today, we only measure those tokens by cost: How many tokens? How much per million? That is like measuring the Eurostar only by speed and cost.

This is the next step for ContextON.AI and ODEFTO AI Labs: Not just to talk about context, but to show it, token by token, in dollars, in time saved, in trust created.

SLCA in practice: A 5-step method to shift context

You do not need permission to do this. You can do it on Monday morning.

Step 1: Name the old C.
Write down the context you are currently in. Example: “We are in a speed context. Success = faster delivery.”

Step 2: Name the aspired V.
Write down the value you actually want, not the metric you inherited. Example: “We want customers who trust us and return.”

Step 3: Ask C = f(Q,V).
If this V is what matters, what C would produce it? List at least 3 different contexts. Do not judge them yet. Example: C1 = Faster, C2 = Clearer communication, C3 = More human support, C4 = Wi-Fi on the train.

Step 4: Test cheap.
Pick the cheapest context to test. Sutherland’s Wi-Fi is cheap. A prototype, a 1-week trial, a single conversation, a small service. Do not build a tunnel before you test Wi-Fi.

Step 5: Keep the loop.
Value and context form a loop, not a line. When you get new V, it reveals a new C. When you try a new C, it reveals a new V. That is why the work never finishes, and why that is good news. The loop is the job.

Apply this to three real places:

For a student: Do not ask, “How do I get better marks in this context?” Ask, “What value do I want to be recognised for? A job? A portfolio? A skill? What context would make that visible?” Then build that context: A project, a teaching video, a small tool.

For a professional: Do not ask, “How do I do my tasks faster with AI?” Ask, “What value does my team actually need? Trust? Clarity? Speed? What context would let AI produce that value, and what context should stay human?”

For a founder: Do not ask, “How do I use AI to cut costs?” Ask, “What value do my customers aspire to? If I start from that V, what C should I build that no AI can copy by being faster?”

In each case, you are not competing with AI on V = f(Q,C). You are leading with C = f(Q,V).

Why this is good news for job creation

The fear is: AI will make the train so fast, no driver is needed.

The reality is: AI makes the old context cheap, so the new context becomes valuable.

When speed becomes free, enjoyment becomes scarce. When answers become free, good questions become scarce. When content becomes free, trust becomes scarce. When execution becomes free, context-shifting becomes scarce.

Jobs will not disappear. Jobs will change form, from doing tasks inside a given context to recognising ignorance, recognising context, and acting to create value.

That is learnable. That is teachable. That is what ODEFTO Learning Labs, ODEFTO Press, and UpSkillShare are building: Not more certificates for the old context, but training in SLCA, IAF, and RAT for the new context.

We do not need more people who can build a faster tunnel. We need more people who can ask: Do we need a tunnel at all? Or do we need Wi-Fi?

Conclusion: Do not ask for a faster train

Rory Sutherland did not hate engineering. He loved psychology enough to notice when engineering was solving the wrong problem.

AI does not hate humans. It loves the context we give it enough to optimise it perfectly, even if it is the wrong context.

The need of the hour is not to fear AI, nor to worship it. It is to learn to shift context on purpose.

Stop asking only: How do we make this faster?

Start asking: What value do we aspire to, and what context would make that value visible?

That is:

Value = f(question, context) : Context = f(question, value)

That is SLCA Engineering.

And that is how we stop spending billions to save 40 minutes, and start spending 0.01% to change a life.


Next step for the reader: Take one task you are doing with AI this week. Write down the old C, the aspired V, and three new Cs that could produce that V. Test the cheapest one. That is SLCA.

Learn more at ODEFTO.COM and THE AI TODAY.

Suggested slug for publishing: stop-making-the-train-faster-shift-context-slca
Suggested meta description: Rory Sutherland’s Eurostar story explains the AI age: AI is brilliant at making the old train faster. Humans must learn to shift context to create value. V = f(Q,C) : C = f(Q,V) and SLCA Engineering.

By Kiran Vadagam, ODEFTO Labs | For THE AI TODAY

Leave a Reply

Your email address will not be published. Required fields are marked *