The new tech leader’s problem

I first wrote this book for new technology leaders in 2021. The problem was easy to describe and hard to solve: business and technology teams needed to work together better.

Business leaders needed to understand how technology created value. Technology leaders needed enough business context to build what mattered, not just what was requested. Product, engineering, design, operations, and executives needed a shared way to talk about purpose, problems, tradeoffs, delivery, and outcomes.

That problem is still here. Now it has agents in the room.

Technology leadership is no longer only about software teams, roadmaps, systems, and delivery. Leaders are also managing workflows that include assistants, coding agents, automations, and agents that act inside company systems. Some draft content. Some analyze data. Some write and ship code. Some summarize meetings. Some connect to the CRM, the ticket board, or the warehouse through MCP servers and call tools on their own. Some hand work to other agents. A few become important before anyone has clearly decided who owns them.

Most organizations are not struggling with AI because the models are incapable. They are struggling because the operating system around the work is unclear. By operating system, I mean the management layer of outcomes, workflows, owners, decision rights, governance, and cadence that lets work move without constant reinvention.

The company has pilots, but no portfolio.

Tools, but no ownership model.

Automation, but no workflow map.

Agents, but no lifecycle.

Enthusiasm, but no review rhythm.

Output, but no accountable owner.

This is the new tech leader’s problem: turn scattered AI activity into governed execution without slowing the company into bureaucracy or letting it drift into chaos.

The old collaboration problem became an operating-system problem

The original thesis still holds: technology value requires business and technical teams to operate together, not align once at kickoff and then throw work over the fence.

The system now has more moving parts. A founder wants faster AI progress. A CTO sees dozens of pilots with no reliability standard. A COO finds five versions of the same workflow. A product leader sees prototypes everywhere and few customer-experience changes. Engineering discovers coding agents opening pull requests faster than reviewers can absorb them.

The issue is not a lack of tools. It is a lack of operating model.

AI does not remove the need for leadership. It increases the cost of unclear leadership. When direction was weak in a software-only organization, teams got rework and delay. In an AI-enabled organization, weak direction produces drafts, tickets, reports, messages, and code at the speed of confusion.

A leader’s job is not to make everyone use AI. It is to make sure AI is used for the right outcomes, inside owned workflows, with decision rights that make accountability visible.

AI sprawl is speed without ownership

Every technology wave creates sprawl. Cloud created infrastructure sprawl. SaaS created tool sprawl. Data created dashboard sprawl. Agile created ritual sprawl. AI is creating agent sprawl.

Agent sprawl appears when teams cannot answer basic questions:

• What outcome should this workflow improve?
• Who owns the workflow?
• What data does it use?
• What systems and tools can it touch, and with whose permissions?
• What decisions can it influence?
• Which other agents does it hand work to or take work from?
• What must a human review?
• How is its output evaluated before and after release?
• What does success look like?
• What failure modes matter?
• How often is performance reviewed?
• When should this agent change, pause, or retire?

These are simple questions with uncomfortable answers. They cut across product, engineering, operations, legal, security, sales, marketing, finance, customer success, and leadership. They reveal unclear ownership, heroic coordination, data fragmentation, and vague definitions of quality.

AI sprawl is rarely one giant failure. It is usually many small missing operating-system parts becoming visible at once.

Humans, workflows, and agents now move together

“AI agent management” can sound technical. Part of it is. The hardest parts are usually managerial.

A useful agent needs more than a prompt or a better model. It needs purpose, context, scoped permissions, a workflow, evals, an accountable owner, and a safe way to fail.

Agents are not employees, magic, or replacements for judgment. They are participants in work.

If an agent drafts customer emails, someone still owns the customer relationship, approved claims, tone, data boundaries, and edge-case review. If a coding agent writes the change and opens the pull request, engineering still owns tests, architecture, merge and deployment controls, and incident response. If an agent analyzes sales calls, revenue leaders still own the decisions and behavior changes that analysis should support.

The work changed. Accountability did not disappear.

Direction before speed

One dangerous myth in AI adoption is that speed solves strategy. It does not. Speed amplifies strategy. It also amplifies confusion.

Before asking “How can we use AI here?” ask:

• What outcome are we trying to improve?
• What workflow creates or blocks that outcome today?
• Where does human judgment matter most?
• Where is work repetitive, rules-based, or context-heavy enough for AI help?
• What would need to be true for this to be reliable?
• Who owns the system after the pilot?

These questions slow the conversation just enough to prevent expensive false starts. They do not make the company anti-AI. They make it serious.

The goal is operating leverage

A company does not become AI-native because it has many agents. It becomes AI-native when AI is integrated into how the company senses, decides, executes, reviews, and improves.

AI-native teams are often small. Agents do much of the drafting, coding, research, and routing. People spend more of their time setting direction, reviewing, and deciding. That only works when the operating system around the agents is clear.

In some companies, maturity will mean fewer agents with clearer ownership. In others, it will mean a larger portfolio with strong lifecycle controls. In most, the first breakthrough will come from mapping the work that already exists and finding the few places where AI can remove friction from high-value workflows.

Operating leverage means better outcomes without proportional coordination cost. It means people can learn faster without drowning in meetings. Leaders can see what is happening without begging for manual status. Agents handle repeatable work while humans focus on judgment, relationships, creative problem-solving, and accountable decisions.

But leverage requires design. Without design, automation becomes another tax.

Chapter 1 operating check

Before moving on, answer:

1. Where is AI already being used inside important workflows?
2. Which AI pilots or agents have clear business owners?
3. Which are producing output without a review or success standard?
4. Where is AI increasing coordination load instead of reducing it?
5. What workflow should be redesigned before another tool is added?
6. What AI-enabled activity should be paused, clarified, or formally owned this month?

If these questions are hard to answer, begin with visibility, not another pilot.

The next step is to understand why technology value has become an operating-system problem.