Direction Over Acceleration
When Execution Becomes Cheap, Strategic Choice Is the Ultimate Competitive Advantage
AI conversations have a natural starting point: What can this technology do?
It is an understandable question. Every few weeks, the answer gets larger. Models can write, analyze, code, search, reason, generate images, operate software, and increasingly take autonomous action rather than merely recommend it. Each new capability triggers another round of demonstrations, experiments, implementation plans, and executive conversations about where AI should go next.
Eventually, leadership needs to ask a fundamentally different question: What progress are we trying to make, and has AI changed the best way to make it?
That sounds simple, but it is not. When a new technology dramatically expands what is possible, capability has a way of becoming its own argument. We discover that something can be automated, accelerated, or delegated, and we begin treating that discovery as evidence that it should be.
But capability does not supply purpose. A modern GPS navigation system can recalculate routes in milliseconds, predict traffic delays, and guide a vehicle with flawless precision. But no matter how sophisticated the software becomes, it cannot choose the destination. It expands the set of reachable places while leaving leadership with the non-delegable problem of deciding where the organization needs to go.
Abundance Moves the Bottleneck
For most of organizational history, execution has been expensive. Writing took time. Analysis took time. Software took time. Research took time. Creating alternatives took time. Even producing a bad first draft consumed enough resources that organizations learned to ration who could create one and what deserved to be created.
Generative AI changes those economics dramatically. Candidate answers, analyses, designs, drafts, plans, and software can now be produced at a cost and speed that would have seemed absurdly optimistic only a few years ago. But when something becomes abundant, the bottleneck does not disappear. It moves, often in unexpected ways.
Prudent AI is an extended response to this migration of scarcity:
- Candidate answers become abundant, so selection, problem framing, and strategic intent matter more.
- Cheap experimentation becomes abundant, so commitment discipline and staged exposure matter more.
- Generated output becomes abundant, so proof, empirical validation, and truth become scarcer.
- Delegable execution becomes abundant, so specification sovereignty and intent control become scarcer.
- Available agency becomes abundant, so warranted authority and reconstitution capacity become scarcer.
When candidate answers become free, knowing which problem is worth solving becomes the rarest asset in the enterprise.
If you can generate ten plausible strategies before lunch, generating strategies is no longer the scarce capability. What is scarce is the ability to choose wisely among them. If software can be built rapidly, the question moves from whether you can build something (feasibility) to whether it should be built (desirability). Everyone can build now, so building is no longer scarce. Knowing what to build is.
This is why I have resisted framing Prudent AI primarily as traditional AI governance. Governance matters, but governance starts far too downstream if leadership has not first decided what progress it is trying to make.
“Implement AI” Is Not a Strategy
This is where Jobs to Be Done (JTBD) becomes essential to my worldview and to yours as well. The Jobs to Be Done framework of Clayton Christensen and others asks us to look past the product or process and ask what progress someone is trying to achieve. That is a vital discipline when technology changes quickly, because new capabilities make inherited processes look far more permanent than they really are.
“Implement AI” is not a job to be done. Neither is “keep up with AI.” Those are responses to a technology. They tell you nothing about the progress the client, customer, professional, or institution is trying to make and would call a success.
Consider a monthly reporting process. A team spends twelve days collecting data, manipulating spreadsheets, writing commentary, formatting charts, reviewing drafts, and assembling a 300-page report. AI arrives, and someone proposes using it to reduce the production cycle from twelve days to three.
That may be a worthwhile improvement, but it is asking the wrong question. Why does the 300-page report exist? What decision is somebody trying to make with it? Which information actually changes that decision? Organizations tend to treat the inherited process as the job. Do not confuse the artifact with the job that caused you to create the artifact.
If the underlying operational data can now be queried continuously, the important AI opportunity is not producing the inherited report nine days faster. It is eliminating most of the report entirely. That is the critical difference between labor substitution and process deletion. One asks how AI can perform existing steps more efficiently; the other asks whether those steps remain necessary once the underlying technological constraint has vanished.
The Skeuomorphic Trap
Much of enterprise AI remains stuck in a skeuomorphic phase: taking workflows engineered around legacy human constraints and simply slapping an AI engine underneath them. We treat AI the way early enterprise software treated paper. We digitized the physical document while keeping the margins, page breaks, and filing cabinets intact. We removed the paper, but preserved all of paper’s operational friction.
An associate reads documents. Now AI reads them. An associate produces the first draft. Now AI produces it. Someone reviews a contract. Now AI reviews it. Real value exists there, but a Jobs to Be Done approach moves the question upstream.
One of my favorite examples from legal AI vendors is using AI to reduce the drafting and negotiation of an NDA down by a substantial percentage of time. Alexander the Great’s Gordian Knot approach to that would be to turn standard NDAs into clickthrough agreements and reduce the time and effort to zero.
What is the client trying to accomplish? Reduce uncertainty enough to sign the transaction? Identify the handful of risks that could materially change the deal? Get to an acceptable allocation of risk quickly?
If the client’s job is not “receive a contract draft” but “reach an acceptable allocation of risk so the transaction can proceed,” then AI changes far more than who drafts the document. It alters what gets surfaced, when negotiation occurs, what gets escalated, and whether parts of the traditional document production sequence remain necessary at all. Instead of sending forty pages of boilerplate back and forth to locate three disputed risk points, an AI-enabled system can compare risk profiles directly and surface only the substantive operational gaps for human negotiation.
Challenge automating the existing assembly line. Focus on accomplishing the real job.
The Direction Filter
The technology keeps making execution less scarce. What becomes infinitely more valuable are the capabilities that tell execution where to go: sensing what has changed, framing the problem, selecting among alternatives, determining what evidence is sufficient, deciding what deserves authority, and recognizing when new evidence demands a change in direction.
Speed needs direction. When experimentation becomes cheap, organizations should become extraordinarily fast at testing assumptions, exposing bad ideas, comparing alternatives, and discovering what deserves commitment.
Getting rapidly to a rough starting point is valuable because a 0.2 version makes an idea inspectable. You can test it, criticize it, compare it with alternatives, discover that you asked the wrong question, and throw it away without having invested heavily in making it polished.
That is useful acceleration because it improves orientation. But acceleration by itself is not progress. Velocity requires speed and direction. The discipline lies in knowing the difference between accelerating learning and accelerating commitment. An organization that becomes dramatically better at executing the wrong process, optimizing the wrong objective, or solving the wrong problem has not become more strategic. It has simply become faster.
Before allocating significant capital, redesigning a workflow, or issuing a mandate to “use AI,” leadership must run every initiative through three forensic questions:
- The Job: What progress are we trying to make for the client or institution?
- The Constraint: What specific friction prevents that progress today?
- The Design: Does AI alter the best way to accomplish the underlying job, or are we merely paying a complexity tax to make an obsolete process run faster?
Notice what is missing. The first question is not, “Where can we use AI?” That question comes later.
Kennedy’s Law and Strategic Choice
The temptation in a period of rapid technological change is to treat speed as the scarce resource. Someone else is moving faster. A competitor announced something. A vendor released a new model. The organization feels panicked pressure to produce an AI strategy, an agent strategy, an implementation roadmap.
Move where? That question is not a brake on progress. This brings us to what I like to playfully call Kennedy’s Law:
Any serious conversation about artificial intelligence drives us quickly and inevitably to the most fundamental questions about an organization, whether we intended to ask them or not.
What began as a tactical discussion about a software trial or prompt engineering rapidly escalates into questions of institutional purpose, core competencies, risk tolerance, decision rights, and what value we deliver. AI acts as a mirror that exposes every ambiguity in your strategy. You cannot decide how to deploy artificial intelligence until you decide what your organization actually exists to accomplish.
Prudent AI is not a prescription for moving slowly. It governs consequential AI commitments under uncertainty while preserving the capacity to learn, validate, adapt, and recover. Move fastest where experiments are cheap, failures are reversible, and learning is valuable. Demand progressively stronger evidence as commitments become harder to unwind and the consequences of error increase.
Prudent AI isn’t ultimately about governing AI. It is about making better consequential choices in a world where AI has radically expanded the available choice set. In an era where acceleration is abundant, strategic direction is the ultimate competitive advantage. We can navigate faster than ever. The satellite routing can recalculate our position in an instant. Leadership still must decide where to drive.
Accelerate reversible learning. Pace irreversible commitment.
Dennis Kennedy – CC BY 4.0 license
[Originally posted on DennisKennedy.Blog (https://www.denniskennedy.com/blog/)]
DennisKennedy.com is the home of the Kennedy Idea Propulsion Laboratory
DennisKennedy.Blog is part of the LexBlog network.
Recent Comments