Series 7 | The Goal Management Paradox: The Better We Get at Achieving Goals, the Farther We May Drift
Subtitle: From an Ancient Chinese Parable to Co-Growth Between Humans and AI Agents
01 When the Direction Is Wrong, the Faster the Horse, the Farther You Go
The Zhanguo Ce (Strategies of the Warring States), a Chinese classic compiled more than two thousand years ago, tells a famous story known as nán yuán běi zhé — literally, "south-pointing shafts, northbound wheel-tracks."
A man is preparing to travel to the State of Chu, which lies to the south, yet he drives his chariot north.
A passerby warns him: "Chu is in the south. You are going the wrong way."
He replies: "No matter. My horse runs fast."
The passerby keeps urging him, and he explains: "I have ample money for the road, and my driver is excellent."
He owns a fine horse, abundant resources, and a superb executor. It seems he has everything success requires.
Except the direction.
The faster the horse, the fuller the purse, the more skilled the driver — the farther he may travel from his destination.
A parable from more than two thousand years ago still gives us pause today.
We have ever more advanced tools, ever more efficient organizations, and AI Agents capable of autonomously carrying out complex tasks.
Yet one question never seems to go away:
As we get better and better at getting things done, do we also find it easier and easier to overlook what we ought to be doing in the first place?
02 Why Do Humans Need Goal Management?
Suppose you are thirsty and there is a glass of water on the desk.
Pick it up. Drink.
The distance between need, action, and result is so short that hardly any goal management is required.
But if your wish is to build a successful company five years from now, everything changes.
You must decide what product to make, which users to serve, how to allocate resources — and when to persist, and when to let go.
The outcome of today's actions may not be known for months, or even years.
And once many more people are involved, each of them holds different information and bears different responsibilities.
So we need goals.
Goals give scattered actions a shared direction, let limited resources be concentrated, and connect a distant future to today's choices.
From the long voyages and city-building of the ancient world to modern corporate strategy, MBO, and OKR, humanity has never stopped developing this capability.
Yet what goal management truly has to solve is not just "how to complete a goal."
Because when we set goals, our understanding of the world is usually incomplete.
Goals can keep us from getting lost, but they cannot guarantee that the direction we chose is right.
03 A Goal Is a Compass — and It Can Also Become a Shackle
Imagine a software startup.
The founder notices that product delivery is slowing down, and announces:
"Over the next three months, we will double our development efficiency."
The team starts optimizing processes and introducing automation tools, and makes the number of tasks completed each month a key metric.
Three months later, the numbers look beautiful.
More commits, faster feature releases — everyone seems more productive.
But users' problems have not decreased, and rework caused by misunderstood requirements keeps increasing.
After investigating, they discover that what was truly slowing the project down was never coding speed — it was that requirements were often not fully understood.
The team successfully optimized a metric without ever solving the original problem.
This brings to mind Goodhart's Law in economics: when a measure becomes a target, it may gradually lose its ability to measure the very thing it was meant to capture.
But the deeper problem goes beyond metrics going astray.
Even if the team really did double its development speed, what would that efficiency be worth if they were building features users do not need?
Goals allow us to concentrate resources, resist distraction, and sustain long-term action.
Yet once we place too much faith in a goal, it can also keep us from accepting new evidence.
The Lüshi Chunqiu, another ancient Chinese classic, tells of a man crossing a river by boat who dropped his sword into the water. He at once cut a notch on the gunwale to mark the spot, planning to search for the sword beneath the notch after landing.
The trouble is, the boat has kept moving forward, while the sword remains where it fell.
The world has already changed, yet he still trusts an old coordinate.
Real organizations sometimes do the same.
We write goals into plans, break plans down into metrics, turn metrics into performance reviews, and then execute day after day.
What we rarely re-examine is this: has the world that the goals were based on already changed?
And so goal management reveals an inner paradox:
Without stable goals, we can hardly accumulate results; but when goals are too stable, we may lose the ability to adapt to reality.
So when should we hold course, and when should we turn?
04 Real Goal Management Is More Than Managing Execution
Cybernetics tells us that a system can adjust its actions through feedback, continually comparing the gap between what it expects and what is real.
Like a thermostat.
Set it to 25°C: heat when the room is colder, stop when it is warmer.
The mechanism works remarkably well.
But the thermostat never asks:
"Why does the room have to stay at 25°C? What if no one is in it anymore?"
This is exactly where the difficulty of complex goal management lies.
In research on organizational learning, Argyris and Schön distinguished single-loop learning from double-loop learning.
Single-loop learning asks:
"How can we do this better?"
Double-loop learning goes on to ask:
"Why do we believe this is worth doing in the first place?"
The former improves methods; the latter re-examines the assumptions and goals that govern our actions.
Both are indispensable.
If we change the goal every time we run into difficulty, we will never accumulate anything.
If we stick to the original goal no matter what evidence appears, we may repeat the same mistakes again and again.
That is why I prefer to understand goal management as:
In an uncertain world, continually coordinating intention, action, and results — and using feedback from reality to decide whether to fix the method, fix the understanding, or re-examine the goal itself.
It is not a list of objectives; it is an ongoing process of learning.
05 When the AI Agent Becomes That Ever-Better Charioteer
Now, let us return to that software company.
Suppose the founder brings in a powerful AI coding Agent.
It can understand requirements, draw up plans, write code, run tests — and even autonomously complete tasks that once required an entire team.
The founder tells it:
"Our goal is to improve development efficiency."
The Agent executes diligently.
It keeps optimizing the process, completing development tasks faster and faster.
As a result, the team may produce features that users never truly needed — several times faster than before.
In a sense, we are manufacturing ever-better charioteers without necessarily possessing an equal ability to judge direction.
But the story could also end another way.
After months of work, the Agent notices that much of the rework happens at the requirements-understanding stage.
So it proposes to the founder:
"Based on the record so far, coding speed may not be the main bottleneck. Should we re-examine our current efficiency goal?"
The founder examines the evidence and finds that the problem really does lie in his own way of working.
He begins to clarify user scenarios and success criteria before development starts.
The Agent adjusts its behavior as well: when facing vague requirements, it no longer rushes to write code, but asks clarifying questions first.
Later, the team verified this change through real delivery outcomes.
This time, it was not just the AI that changed.
The human's understanding changed, and the Agent's behavior changed too.
Through feedback from the real world, the two sides formed a new way of working together.
Of course, an Agent being able to make a proposal does not mean it may decide goals on its own. Who has the authority to change direction, what evidence is required, and who bears the consequences — these questions still must be answered.
And one successful suggestion does not prove that co-growth has already happened. Real learning has to be repeatedly validated in future actions.
Yet this possibility alone is worth thinking about.
06 From Completing Tasks to Growing Together
In the past, we mostly cared about making our tools more powerful.
Today, we are beginning to possess agents that can understand language, draw up plans, and act autonomously.
This makes an old question urgent again:
When the capacity to act grows faster than the capacity to judge, how do we keep from marching ever more efficiently toward the wrong future?
Perhaps human–AI collaboration cannot stay forever at:
humans set the goals, AI carries out the execution.
We also need to explore a new kind of relationship:
humans provide the direction of value and the authorization; agents provide cognition and the capacity to act; and both sides, through feedback from the real world, keep correcting their understanding of the problem and improving their future behavior.
This does not mean AI should decide the direction of a human life, nor that every task requires its goals to be debated anew.
Rather, as long-term collaboration accumulates experience, the system should become able to discover: which methods need to change, which assumptions may be wrong, and which goals no longer fit reality.
Return, once more, to the story of the southward journey aboard the northbound chariot.
If the traveler in that story had an AI charioteer today, we would of course want it to have the fastest horse, the best navigation, and the finest driving skill.
But perhaps what we would want even more is for it to say, when it truly matters:
"Are we going the wrong way?"
And a human–AI system truly worthy of trust should not only be able to raise this question, but also help us search for evidence, make the judgment, and carry this lesson into the journeys ahead.
After all, what has ever mattered is not just how fast we travel.
It is whether, over the long road, we can keep learning how to tell the direction.
Next: A Goal Is Not a Prompt: When AI Begins to Act Autonomously, How Should We Govern Goals Themselves?