The jagged frontier: Using AI without losing your edge

The adoption reality

McKinsey’s 2025 global survey reports that 88% of respondents say their organizations are using AI in at least one business function, up from 78% a year earlier. At the same time, nearly two-thirds say their organizations have not yet begun scaling AI across the enterprise, which means most teams are still experimenting rather than compounding results. McKinsey also notes rising interest in agentic AI: 62% of respondents say their organizations are at least experimenting with AI agents, yet agent scaling remains limited in most functions.

This pattern creates a familiar leadership problem: tools are everywhere, but outcomes are uneven. If AI is treated like a shortcut, it becomes noise; if it’s treated like infrastructure, it becomes leverage.

The “jagged frontier” problem

A major reason results vary is that AI performance isn’t smooth, it’s “jagged.” In a study summarized by MIT Sloan (based on research with more than 700 BCG consultants), using GPT‑4 on tasks “inside the frontier” increased performance versus no-AI conditions, including a reported 38% lift for GPT-only and a 42.5% lift when participants also received an overview on how to use it. But when participants used AI on tasks “outside the frontier,” performance dropped, including decreases reported as 13 percentage points (GPT-only) and 24 percentage points (GPT plus overview).

The leadership takeaway is uncomfortable but liberating: AI doesn’t replace judgment; it changes where judgment is most needed. As the researchers observed, performance can decline when people “switch off their brains and follow what AI recommends,” especially when the AI sounds confident while being wrong.

The hidden tradeoff: Motivation

Even when AI helps, it can introduce a new kind of drag. Harvard Business Review summarizes research (over 3,500 people) finding that gen AI use can increase performance, but also reduce intrinsic motivation and increase boredom when people return to tasks where AI isn’t available. In other words: AI can boost output today while quietly changing how people feel about effort tomorrow.

For founders and senior operators, this matters because motivation is a strategic asset. If the team becomes dependent on AI for the “interesting” parts of work, leaders may get short-term speed at the cost of long-term craft.

A simple operating system for leaders (and coaches)

McKinsey reports that 51% of respondents from organizations using AI have seen at least one negative consequence, with nearly one-third of all respondents reporting consequences stemming from AI inaccuracy. So the goal isn’t to “use AI more”, it’s to use it with rules that protect quality, confidentiality, and credibility.

Here’s a clean model to run:

• Use AI for acceleration, not authority. The MIT Sloan/BCG findings show AI can meaningfully boost performance on the right tasks, but can reduce performance on the wrong ones. Treat outputs as drafts, options, or checklists, never as final decisions for high-stakes calls.

• Define “inside-the-frontier” work. The study’s pattern implies leaders should explicitly identify tasks where AI reliably helps (e.g., first drafts, summarizing, reformatting, ideation) versus tasks that require deep contextual reasoning and verification. When the task sits near the edge, add friction: second-pass human review, source checks, and a decision memo written in plain language.

• Redesign workflows, not just prompts. McKinsey emphasizes that redesigning workflows is a key success factor for high performers, and most organizations have not embedded AI deeply enough to see enterprise-level benefit. In practice, that means deciding where AI lives (CRM, inbox triage, content pipeline, support desk), who owns it, and what “good” looks like in measurable terms.

• Keep ethics and confidentiality explicit. The International Coaching Federation’s AI Coaching Framework and Standards highlights both opportunities (like accessibility and efficiency) and risks including bias and confidentiality, and it frames standards across domains that include foundational ethics and technical factors like privacy. For coaching businesses, that translates into simple policies: what client data can be used, where it can be stored, and what must never enter a public model.

• Protect human meaning. HBR’s summary that AI can reduce motivation suggests leaders should intentionally preserve “human zones”: strategy, relationship, and craft, work that builds pride and identity. If AI handles the first 80%, make the final 20% the place where taste, standards, and accountability are trained.

If this becomes the norm, AI turns into something calmer than hype: a reliable assistant that reduces cognitive load while strengthening decision quality. That’s the “modern wisdom” move, use machines for speed, and humans for judgment.

Sources

1. McKinsey & Company, “The State of AI: Global Survey 2025” (AI usage, scaling, agents, and risk/negative consequences).

2. MIT Sloan Management Review (Ideas Made to Matter), “How generative AI can boost highly skilled workers’ productivity” (BCG/GPT‑4 study results; jagged frontier).

3. Harvard Business Review, “Research: Gen AI Makes People More Productive—and Less Motivated” (productivity vs motivation tradeoff).

4. International Coaching Federation (ICF), “AI Coaching: Framework and Standards” (ethics, confidentiality, privacy domains).