Measure Twice, Automate Once.

The Contractor Problem

I spend a great deal of time thinking about how context moves between a person and a machine. My work focuses on frameworks that keep both implicit and explicit details visible: what the user actually said, what the system inferred, what assumptions are being made, and where uncertainty remains.

The purpose is not bureaucratic documentation for its own sake. It is to let the user inspect the work while it is still being formed, so that a small misunderstanding can be corrected before it compounds into a completely wrong result.

The most useful analogy may not come from computer science at all. It comes from home improvement.

Imagine hiring a contractor to renovate part of your house. You know what you want in a general sense, but much of the picture exists only in your head. You assume certain things are obvious: how you use the room, what you want preserved, what compromises you are willing to accept, and what would make the finished work unusable.

The contractor, meanwhile, brings a different set of assumptions shaped by previous jobs, personal preferences, available materials, time pressure, and habit.

If you provide almost no context and finally say, “Just do it however you want,” you have not eliminated the need for decisions. You have merely transferred those decisions to someone whose mental model may be entirely different from yours.

There is a Seinfeld episode built around exactly this problem. Jerry becomes frustrated with the contractor renovating his kitchen cabinets and effectively gives him permission to do whatever he wants. The result is bizarre, intrusive, and fundamentally incompatible with how Jerry lives.¹ The contractor completed the work. He followed the freedom he was given. Yet the outcome was still wrong because the real requirements were never successfully shared.

That is what happens when a person gives an artificial intelligence an underspecified task and assumes the system will somehow reconstruct the intended result. The machine does not stop making assumptions when context is missing. It fills the empty space with defaults.

Missing context does not eliminate decisions. It transfers them.

This is why the old home-improvement maxim remains so powerful:

Measure twice, cut once.

It sounds like advice about lumber, but it is really a philosophy of action. Inspection should precede commitment. You should verify the model before making the irreversible cut. A measurement error that costs ten seconds to correct before cutting may cost hours, money, and material afterward.

The same principle applies to software, technical support, AI systems, organizational decisions, public policy, and almost every other form of consequential work. The earlier an error becomes visible, the cheaper it is to correct.

Context-management frameworks are therefore not simply methods for producing better text. They are the cognitive equivalent of checking the measurement before the saw touches the board.

The Education Hidden Inside Home-Improvement Television

The home-improvement programs of the 1980s and 1990s were quietly teaching more than construction.

The real programs, such as This Old House, showed people how physical systems fit together. The fictional programs built around that culture, especially Home Improvement, turned the same principles into comedy.

Tim Taylor wanted more power, bigger machinery, and faster results. Al Borland understood materials, procedures, and limits. Tim was fascinated by the tool’s potential. Al understood the conditions under which the tool could be trusted.

That distinction is more important than it first appears.

A tool is not valuable merely because it is powerful. Its real value depends on the depth of the user’s understanding. One tool understood thoroughly may be more useful than a workshop filled with equipment the owner barely knows how to operate.

Familiarity creates judgment:

• when to use the tool;
• when not to use it;
• what failure sounds like;
• what resistance feels normal;
• and what signs indicate that the process should stop.

Modern technology encourages the opposite relationship. We are surrounded by systems that conceal their operation and promise useful outcomes with minimal understanding. The interface invites us to press a button and accept whatever happens.

When the result is good, the abstraction appears magical. When it fails, the user often has no model for understanding why.

The tragedy is not merely that systems make mistakes. It is that users are increasingly trained to believe understanding is unnecessary.

Home-improvement culture assumed that competence could be taught. Even viewers who would never rebuild a staircase could learn that structures have load paths, materials have tolerances, and shortcuts have consequences.

The physical world pushed back. A beam either carried the load or it did not. Water followed gravity regardless of the homeowner’s opinion. Electrical current did not negotiate with confidence.

A builder develops trust in the physical world because that trust is grounded in repeated contact with reality. They know where a beam must go because they understand what the beam is doing. They do not need to treat the structure as a mystery, a conspiracy, or an act of faith. The system can be inspected.

That kind of grounded trust is increasingly difficult to maintain in digital life.

The Monolith, the Bone, and the Meaning of a Tool

The opening movement of 2001: A Space Odyssey presents one of cinema’s most enduring images of technological transformation. The monolith appears, and the apes begin to understand the bone as a tool.

The bone is not simply another object in the environment. It changes the range of actions that can be imagined.

The tool extends the mind.

Once the bone can be understood as a weapon, a lever, or an instrument, the environment itself appears different. Possibilities that were previously invisible become actionable.

The famous cut from the bone thrown into the air to a spacecraft orbiting Earth compresses the history of technology into a single transition. The spacecraft is still the bone. It is matter reorganized by intelligence to extend what intelligence can do.

Artificial intelligence belongs to that lineage. It is not merely another product category or productivity application. It can become part of the user’s thinking loop. It helps organize information, test interpretations, generate alternatives, expose patterns, and convert vague intention into executable form.

But every extension of capability also extends the consequences of misunderstanding.

The bone allows the ape to do more, but it does not make the ape wiser. The power of the tool and the judgment of its user are separate variables. A machine that amplifies reasoning can amplify confusion, bias, carelessness, and false confidence just as efficiently as it amplifies insight.

The proper response is not to reject the tool. Nor is it to surrender judgment to it. The response is to construct a relationship in which the user can see how intention becomes action.

That is where context becomes essential.

Context is not background decoration. It is the load-bearing structure of collaboration.

Support Should Produce Understanding, Not Dependence

My perspective is shaped by a lifetime in technical support. Support work is often described as fixing technology, but the deeper task is understanding how another person understands technology.

A user rarely reports a problem in the language of its actual cause. They report what they can see: the application will not open, the password stopped working, the printer disappeared, the network is slow, or the file is gone.

The technician has to reconstruct the mental model behind the report:

• What does the user believe happened?
• What have they already tried?
• What distinction are they missing?
• What sequence of actions produced the current state?

The easiest form of support is to correct the immediate result. Reset the setting, reconnect the device, restore the file, and close the ticket.

The better form of support identifies what caused the problem and helps prevent it from recurring.

That does not mean every user needs a lecture or should become a systems engineer. It means the intervention should leave the person with a more accurate model than they had before.

Good support restores the system. Excellent support increases the user’s agency.

This creates an important distinction between two kinds of technological assistance:

Both may feel helpful in the moment. Only one compounds in the user’s favor.

A support technician who teaches a transferable principle may reduce the number of future tickets. From a narrow engagement perspective, that could look like a failure: the user returns less often.

From a human perspective, it is success. The person can now recognize the pattern, correct the issue, or ask a better question next time.

Many contemporary platforms are optimized in the opposite direction. Their success depends on keeping the person inside the loop for as long as possible.

The Optimization of Frictionless Dependence

The cultural change did not begin with COVID-19, but the pandemic accelerated it dramatically.

Work, friendship, entertainment, education, commerce, and identity were pushed deeper into digital platforms. Behaviors that had once seemed excessive became normal through necessity. At the same time, platforms gained an unprecedented opportunity to measure, optimize, and monetize those behaviors.

Before that acceleration, online culture already contained manipulative incentives, but there remained stronger social resistance to them. Clickbait was widely treated as embarrassing. Obvious engagement farming damaged credibility. Many creators wanted attention, but they also wanted to produce something that justified the attention.

Communities such as early Reddit were imperfect, sometimes hostile, and frequently self-important, but they still maintained an aspiration toward substantive discussion. The culture distinguished itself from pure chaos by imagining that discourse, evidence, and shared understanding mattered.

Over time, the economic logic of the platform became the cultural logic of the user.

A revealing example is the automated handoff between livestreams. When one creator ends a broadcast with hundreds of viewers, the platform may attempt to redirect those viewers into another active stream. From the platform’s perspective, this is elegant. It removes the moment of friction in which a viewer might decide to close the application.

The viewer does not need to choose what to watch next. The platform chooses on their behalf.

A creator looking at this system may immediately ask:

How can I capture that traffic?
How can I monetize the handoff?
How can I position my stream to receive those viewers?

Those are rational questions inside the incentive structure. But they avoid the larger question:

What kind of human behavior is the system exploiting?

The business opportunity depends partly on weakened attention, habitual consumption, and the removal of conscious stopping points. It converts the difficulty of disengaging into revenue.

What might once have produced moral discomfort is reframed as growth strategy.

COVID did not invent selfishness, opportunism, or algorithmic manipulation. It helped normalize a world in which nearly every behavior could be treated as a measurable funnel.

Once a habit could be optimized, it could be monetized. Once it could be monetized, resisting it could be portrayed as naïve.

The result is a society increasingly weighted toward systems that ask people to accept outcomes without understanding the process. Responsibility becomes strangely fragmented. The platform designs the incentives. The creator follows them. The audience responds to them.

When the outcome becomes harmful, every participant can point somewhere else.

The Accountability Gap

Automation creates an accountability problem when the person initiating the process believes that delegation eliminates responsibility.

It does not.

When a homeowner tells a contractor to do whatever he wants, the contractor remains responsible for professional judgment. But the homeowner has also surrendered part of the decision-making process without ensuring that the contractor understands the intended outcome.

When a creator builds content around compulsive engagement, the platform’s incentives matter. But the creator is not transformed into a passive object with no moral agency.

When a user gives an AI a vague instruction and receives an unsuitable result, the platform may be responsible for overstating the system’s reliability. Yet the user must also understand that delegation without inspection is not the same as collaboration.

Responsibility can be distributed without disappearing.

This is uncomfortable because modern product design frequently promises the opposite. It markets convenience as freedom from involvement. The ideal experience is presented as effortless, invisible, and automatic. The less the user understands, the smoother the product appears.

That model works until something goes wrong.

At that point, the user discovers that they were expected to exercise judgment they were never encouraged to develop.

Content That Leaves the Audience Stronger

There is still room to make good content for its own sake. But it requires choosing a different definition of success.

Content optimized for immediate reach often rewards novelty, emotional intensity, outrage, simplification, and repetition.

Content optimized for human capability may grow more slowly because it asks more from the audience. It may require attention rather than merely capturing it.

But the two kinds of content do not have equal effects.

Good content performs work on behalf of the person receiving it. It gives them language, distinctions, historical context, practical understanding, or a more accurate model of reality.

Those benefits propagate beyond the original interaction:

• A viewer who learns how a structure carries weight may recognize unsafe work later.
• A user who learns why a technical problem occurred may prevent it on another device.
• A reader who gains a better framework for evaluating claims may carry that framework into politics, health, work, and relationships.

This is a form of compounding value.

Extractive content also compounds, but in the opposite direction. It strengthens habits of reaction, passive consumption, and dependence. It may produce impressive metrics while leaving the audience less capable of directing its own attention.

The real divide is not between profitable and unprofitable content. It is between systems that increase agency and systems that consume it.

Measure Twice, Automate Once

The contractor, the home-improvement show, the support desk, the monolith, the livestream algorithm, and the AI assistant all belong to the same argument.

Tools extend human capability, but they also create a new obligation to understand the relationship between intention and outcome.

A trustworthy system should not merely produce an answer. It should help the user see what the answer rests upon.

Assumptions should be inspectable. Context should remain visible. Uncertainty should not be silently translated into confidence. The user should be able to intervene before the metaphorical cabinet has been rebuilt into something unusable.

This is not an argument against abstraction. Civilization depends on abstraction. No person can personally inspect every transistor, supply chain, algorithm, bridge, medication, or institution upon which daily life relies.

It is an argument for calibrated abstraction.

We need enough visibility to understand what has been delegated, what remains our responsibility, and where failure can enter the process. We need tools that do not merely remove effort, but direct effort toward the parts that still require human judgment.

The goal should not be to make people perform every operation manually. It should be to ensure that automation does not sever them from causality.

A good tool helps me do more.
A better tool helps me understand more.
The best tool does both while leaving me more capable of acting without it.

That is the standard I want to build toward: technology that treats the human being not as a source of engagement, a collection of monetizable habits, or an obstacle to complete automation, but as an active participant whose understanding matters.

“Measure twice, cut once” was never merely advice about carpentry.

For the age of artificial intelligence, perhaps the principle should be expanded:

Closing Thesis

The purpose of a tool should not be to remove the human from the loop. It should be to make the human more capable within it—and more capable after leaving it.