The Last Mile of Human Taste: Why Replacing Work With AI Does Not Mean People Disappear

04 May The Last Mile of Human Taste: Why Replacing Work With AI Does Not Mean People Disappear

In my previous piece (Stop building software, start replacing work), I argued that the defining shift of our era is the migration of global service budgets to AI-native systems. I described how the next €100B giants will scale not by selling tools that assist humans, but by building systems—“Harnesses”—that directly deliver completed Outcomes.

The signal from the market has been unambiguous: when a system can absorb entire workflows, the magnitude of revenue is orders of magnitude higher (really) than SaaS.

However, a dangerous misconception is taking root that conflates “outcome delivery” with the “complete replacement of the human.” This reasoning is fatally flawed. Successful outcome-centric companies will not aim to replace work indiscriminately. They will eliminate algorithmic complexity entirely, reallocating human capital to where real value is created. Because, even when AI boosts transactional productivity by 10x, you are still left with crucial bottlenecks that AI cannot solve.

1) Rules vs. Judgment: The Fundamental Divide

To engineer a reliable harness (see the above mentioned article for a development of the concept), we must differentiate between two fundamental categories of labor:

• Rules-Based Work (Algorithmic): Work that can be codified and executed via a flowchart. If a task is sufficiently structured and can be described, AI can—and should—fully execute it.
• Judgment/accountability-Based Work (Intuition): Work that relies on accountability under uncertainty, context, taste/preference-setting or unique human values trade-offs under ambiguity (things that are typically hard to describe). AI’s role here is facilitation, not replacement.

2) Concrete Examples: Where the Harness Meets the Wall

To understand the “Harness,” we must look at where the “Rules” end and “Judgment” begins.

Example A: Software Engineering

• The Rules (AI-Driven): Writing a boilerplate API, refactoring a legacy function, or generating unit tests. These are predictable patterns. An AI Harness can ingest a ticket and deliver the code autonomously.
• The Judgment (Human-Led): “Should we even build this feature? Does this architecture create technical debt that will kill us in two years? Does the UI ‘feel’ right for our specific user base?”
• The Bottleneck: The human manager reviews the AI’s 1,000 lines of code not for syntax (the AI handled that), but for intent and alignment.

Example B: Corporate Legal Review

• The Rules (AI-Driven): Scanning 5,000 documents for a specific “change of control” clause or flagging non-standard indemnity language.
• The Judgment (Human-Led): “Given our current relationship with this vendor and our appetite for risk this quarter, is this specific liability cap acceptable?”
• The Bottleneck: The AI flags the risk; the human manager decides if the company is willing to take the risk.

3) The Organizational Law: Who Manages the Bots?

The promise of AI is massive scalability. An AI harness can boost the throughput of a legal review or a coding pipeline by 10x. But this brings us to an unavoidable organizational reality. Even with massive productivity gains and a constantly moving target, you will always be left with that final 1 to 20% of cases that the rules-based system cannot handle. And their number will naturally explode because the bandwidth of agentic systems is just limited by your budget. These are the bottlenecks.

A common misconception is that people will be let go ‘en masse’ and top management will become the grand “orchestrator,” managing a million AI agents from a single dashboard. Except for a few startups, this is operationally impossible in any company of a certain size. Organizational complexity has not disappeared; it has simply shifted from managing human output to managing bottleneck decisions.

The company’s Management Team cannot review 10,000 code deployments or 5,000 contracts. You need Managers in charge of making judgements and ultimately take accountability. Their job is roughly the same as before (Marketing, Sales, Legal, Engineering, etc) but their function evolves and climbs up the intellectual ladder to:

1. Set the Company’s Tone : Every company has its own DNA, an identity that is defined by the internal values it goes by. Output must comply with the company’s culture.
2. Handle Exceptions: Manage the high-risk, high-context work that the AI correctly identifies as “beyond the rules.”
3. Maintain Accountability: You cannot hold a probabilistic model accountable for a catastrophic error across the thousands or millions of tasks you undertake. A real person must own the outcome when it exits the bottleneck.
4. Prevent Decay: Human review stops “model drift”—the degradation of quality that happens when AI systems iterate without being grounded in real-world human taste.

4) Engineering the Handover 💎

This organizational reality is where the durable moat lives. The winners are building harnesses that don’t just “do the work,” but explicitly manage the handover between the AI engine and the Judgment Manager.

The best systems will:

• Detect the Bottleneck: Identify exactly when a case requires a human “taste” check.
• Synthesize for Speed: Instead of giving the manager more data, the system provides the context needed to make a judgment call in seconds.

5) The Reinvestment Thesis: Building Moats Nobody Thought Possible

The inevitable optimization by AI creates a surplus of human time. For many average organizations, this moment will be viewed as a cost-cutting victory, leading to immediate headcount reduction. For the most competitive and ambitious firms, however, this free time is a high-octane fuel for market conquest.

In a hyper-competitive landscape, human resources are not static line items; they are capital to be allocated. The time previously dedicated to rules-based management of thousands of clerical, legal, or software development sub-tasks can now be radically reinvested in work that could never be done or was previously defined as “non-urgent” or “too expensive.”

Firms building on an AI “harness” don’t just achieve efficiency; they gain an unprecedented capacity for competitive leaps. When they absorb rules-based complexity, they unlock their leaders to:

1. Dramatically Strengthen Moats: Investing time in deep, high-judgment relationship building, strategic partnerships, and structural barriers that a general-purpose AI cannot replicate.
2. Accelerate Product Leaps: Executing foundational research, R&D experiments, and architectural pivots that require immense intuitive risk—work that was never possible when resources were bogged down in day-to-day code maintenance.
3. Deliver Unfair Client Value: Shifting from standard reactive support to deep, proactive client-specific strategic advice, using the AI harness as the data-synthesis foundation for specialized human consulting.

This is especially true in a resource-constrained competitive environment. The AI harness eliminates the “average” work, and the remaining 10% of high-impact human work becomes the primary theatre of competition. The leaders don’t seek a rest when the bots take over; they immediately pivot to building the next, high-judgment layer of client value that separates them from the chasing pack. Their advantage isn’t better AI—it’s how they utilize the human bandwidth the AI unlocked.

Final Thought

We are entering a phase where companies are valued in proportion to the operational complexity they absorb (aka, the extent of the work they are able to deliver), but differentiated by the strategic leaps they make with the resulting human surplus.

The opportunity is not to build better productivity tools. It is to build integrated, outcome-centric systems capable of digesting massive structural burden and intelligently facilitating crucial human insight at every bottleneck.

The next highly valuable companies won’t just be “AI-first”; they will actively be incorporating judgment and accountability into their product design. They will deploy a massive AI harness for one definitive purpose: to ensure their most expensive and unique competitive resource—human talent—is never again wasted on a task that can be described by a rule, freeing them to build client-centric value propositions that are impossible to automate away.