Beyond the Nearest Peak

Breaking the Gravity of Local Maxima

LLMs collapse the cost of exploring alternatives, so leadership quality now hinges on how well we search, not how well we craft. Search is becoming the spine of design and strategy work.

The hardest part of design has never been coming up with ideas. It is letting go of the first workable idea to look for better ones. Historically, we stayed on local maxima—peaks that were fine for survival but too low for excellence—because moving off them was expensive. Rewrites, refactors, and architectural shifts burned real political and temporal capital.

LLMs change that cost structure. We can now explore multiple paths in minutes. Sunk cost is thankfully losing its grip. The work of leadership becomes steering the search rather than defending early choices.

The Landscape of Expensive Mistakes

Every design problem creates a landscape. It is multi-dimensional, full of peaks that represent working solutions and valleys that represent broken builds or failed bets.

In the pre-LLM era, moving across this landscape was punishingly expensive. Switching approaches meant throwing away days or weeks of work. Prototypes were slow to build and painful to discard. Because we couldn’t afford to explore the valleys, we rationally optimized for safety. We picked the nearest promising hill and defended it, even if a Mount Everest of a solution was within sight.

To escape a local maximum in that world, you needed skill stacking. You needed a rare individual who possessed deep domain breadth, knew obscure analogies from other fields, and had the metis — the tacit, situational competence — to guess which dark valley was worth crossing. This was the hero model of design: reliance on the intuition of the few because the empirical testing of the many was too slow.

The Shift: From Climbing to Tunneling

LLMs do not replace that intuition, nor do they replace the need for deep skill. What they change is the physics of movement.

Formally: they collapse the cost structure of exploration. In the past, testing a radically different architecture required building it. Now, you can sample five divergent strategies in an afternoon. This allows for a kind of “tunneling” effect: you can visualize a solution on a distant peak without having to painstakingly hike through the budget and org-debt valley in between.

This changes the texture of the work. Higher-level abstraction becomes a working medium; you iterate on intent and constraints rather than syntax. The loop of propose → implement → critique → revise — which used to take days or weeks — now can happen in minutes or hours. Architectural alternatives emerge early, while the cost of change is low, rather than late, when the concrete has set.

Crucially, none of this uncovers a single “right” answer. There is no final, perfect plan hiding in the landscape — only better and worse bets, discovered through faster, cheaper learning.

But this speed comes with a caveat that many teams miss.

The Virtue of Shallow Breadth

The mistake most teams make is assuming LLMs provides depth. They do not. LLMs provide shallow breadth. They offer typical patterns, first approximations, cross-domain hints, and quick critiques. They are not masters; they are infinite interns.

But this shallowness is valuable: It is exploratory fuel.

The new productive loop is to use that shallow breadth to generate options, and then apply human judgment to select which one deserves depth. You prompt for ten approaches, scan them for viability, and discard nine. This is the “Shallow → Score → Select → Deepen” protocol.

Depth still emerges, but it emerges through guided iteration and human taste. The model accelerates the work of discovering what is worth deepening.

When the Cost Collapses, the Target Changes

When you accept that exploration is cheap, you must accept that your old behaviors are now irrational.

Most executives are using LLMs, if at all, to run the old maze faster—treating AI as a typing accelerator for their teams inside a linear “waterfall” process. This is a category error. The shift isn’t about speed; it’s about reach.

First, clinging to early decisions is now professional negligence. We used to defend our first strategic draft because changing direction was a nightmare. Now that artifacts are cheap, rewriting is cheap. Defending a mediocre solution is no longer prudence; it is vanity. If your organization can test five architectures or approaches in a couple of days, settling for the first one is a failure of rigor—especially when the goal is not to find the answer, but to learn which bets survive contact with reality.

Second, cycle time is the definition of capability. In a world of cheap exploration, the leadership team that tests five divergent paths by lunch has a qualitative advantage over the team that spends the day overfitting to one. Slow exploration is not cautious; it is risky.

Finally, the “safety” of the local maximum is disappearing. Exploitation—improving what you already have—used to be safer than exploration. But when the cost of trying new paths is so low, staying on a local peak “to be safe” is actually the dangerous path. The only expensive mistake left is refusing to try.

Acting as Editors, Not Crafters

This shift demands a fundamental change in identity. We must stop acting like crafters and start acting like editors.

The crafter mindset honors the labor of placing every brick. The editor mindset honors the judgment of choosing which walls are worth building at all. Leverage at the executive level requires redesigning your leadership workflow around search management:

• Shift from hero to orchestrator. Your job is not to champion a single solution. Your job is to insist on multiple variations, score them against constraints, and ruthlessly prune the failures. You are not the genius at the whiteboard; you are the filter for the organization’s attention.
• Define the scoring function. Breadth without judgment is noise. To explore cheaply, you must evaluate quickly. Is the constraint runway? Regulatory risk? Strategic positioning? Exploration without a clear scoring function is drift you can’t afford.
• Prune aggressively. Cheap output creates a new form of debt: the cognitive load of options and initiatives. The primary skill of the AI-augmented leader is not adding more documents, decks, or projects, but shutting down dead ends before they consume headcount and political capital.

What Good Looks Like

How do you actually do this as an executive? You replace the Slot Machine Workflow with the Tree Search Workflow.

The Anti-Pattern: The Slot Machine (Bad)

Too many treat the LLM like a slot machine. They toss in a prompt (“Write a strategy for entering the SMB market”), pull the lever, and accept whatever comes out. If it feels off, they tweak language around the edges instead of questioning the direction. This is high-friction and low-leverage. You are still walking the landscape linearly, just slightly faster.

The Pattern: The Tree Search (Good)

The editorial workflow for leaders looks like this:

• The Fan-Out (Minutes 0–10). Do not ask for a plan. Ask for options. “Propose five distinct strategies for entering the SMB market: partnerships, self-serve product-led, inside sales, ecosystem integrations, and a hybrid. For each, outline benefits, risks, time-to-impact, and required org changes.”
• The Audit (Minutes 10–20). Read the critique. Apply your constraints. Kill the options that don’t fit your runway, risk appetite, or brand.
• The Scaffold (Minutes 20–30). Now, deepen only the survivors. Ask the model to draft just the scaffolds—high-level roadmaps, org charts, and investment theses—for the one or two options that remain.
• The Deepen (Hours 1–4). This is where you take over. You have skipped the “valley” of weak strategies without burning cycles on them. Now you use human leadership—your metis—to pressure-test with your team, model the P&L, surface execution risks, and negotiate trade-offs the model cannot see.

The Difference

The Slot Machine leader spends weeks selling and polishing the wrong strategy. The Tree Search leader spends that time exploring viable paths, learning and iterating through cheap experiments and simulations.

Do Less, Judge More

Licklider foresaw this partnership decades ago: humans supplying the judgment, machines supplying the options. We have finally arrived at that moment, but arriving there requires a violent adaptation in how we value leadership work.

This shifts the seat of our competence. The era of the solitary genius crafting the perfect strategy in one go is over. The era of the curator—scanning a dozen possible futures to choose the next experiment—has begun.

To survive this, we must adapt. We must invert the ratio of our effort: Do less. Judge more. Stop measuring your value by the volume of activity. Measure it by the learning you unlock and the paths you reject.

What “Judge More” Looks Like in Practice

Good judgment is not heroic intuition. It is a steady discipline built on three commitments that fit the new economics of exploration:

• Do not settle on the first workable option A workable idea only shows that one path exists, not that it is the best one. Generate contrasting alternatives before choosing. Treat the earliest idea as a reference point, not a winner.
• Score options before deepening them. Depth is costly. Before investing, compare options against the same constraints: runway, risk, time-to-impact, differentiation, and feasibility. Let the scoring filter the field so depth lands where it matters.
• Keep cycle time visible and short. Exploration speed is now a competitive advantage. Track how long it takes to move from question to options, from options to selection, and from selection to a testable draft. Short cycles prevent overcommitment to weak paths and create continual learning.

These practices give “judge more” a concrete shape. They keep attention on what leadership now optimizes for: quality of search, strength of pruning, and clarity of decision.

A Portfolio Lens on Search

Another way to see this discipline is through portfolio theory. Options are your deal flow. Scoring is diligence. Deepening is capital allocation. Pruning is cutting losses early before they compound. The aim is not to back the first workable idea, but to maintain a healthy pipeline of alternatives and invest only in the ones that survive the filter. This framing reinforces the shift: leadership becomes the management of search quality and portfolio health, not the defense of early drafts.

The machines build. We choose.

---

Appendices

Appendix A: The Topology of "Tunneling" (Complexity)

In Stuart Kauffman’s NK Model, a “rugged” fitness landscape traps agents on local maxima. To find a higher peak, an agent usually has to step “downhill” (worse performance) to cross a valley.

• Pre-LLM: Crossing the valley required expensive refactoring. The cost of the “downhill” move often exceeded the team’s political or temporal capital, trapping them on mediocre hills.
• Post-LLM: LLMs allow for “tunneling.” Because we can generate a fully formed alternative architecture in parallel, we can assess the potential of a distant peak without painstakingly traversing the valley of broken code in between.

The Shift: We no longer walk the landscape linearly; we sample it discontinuously.

Appendix B: The "Shallow-to-Deep" Protocol (Cognition)

The Shallow-to-Deep protocol is the cognitive version of tree search. It widens the branch factor early, applies a simple evaluation function, and prunes most of the space before committing depth to the remaining branches.

1. Explosion (shallow): Prompt for N divergent approaches; maximize variance, ignore syntax errors.
2. Selection (the filter): Apply the scoring function (cycle time, modularity, correctness); reject N−1 options immediately.
3. Deepening (the investment): Inject metis—tacit knowledge, constraints, edge cases—into the surviving option.
4. Verification: Run it. If it fails, do not debug linearly; loop back to step 1 with new constraints.

Key Insight: Do not spend “Deepening” effort until “Selection” has occurred.

Appendix C: The Inequality of Verification (Economics)

The economic shift is simple: the cost of generation collapses, the cost of verification persists. Writing variants is cheap; reading, testing, and rejecting them still takes human time. The bottleneck moves from producing to judging.

• Old rationality: productivity was lines written because writing was scarce.
• New rationality: productivity is bad paths rejected because judgment is scarce.

David Deutsch’s lens applies: progress is the rate at which we correct errors. Faster generation only helps if we raise our rejection and correction rate to match.