Efficiency Is a Dead End

The Local Maximum

AI and efficiency feel like a natural fit. But I’m here to argue that efficiency is a dead end. Chasing efficiency is not a direction. Every company wants to speed things up with AI. Their hope is that order of magnitude results AI promises will result from marginal changes. This is not a strategy.

The aperture is too narrow. Most AI efforts today optimize individual tasks inside broken workflows, making each leg of a relay race faster while ignoring that the race itself is the problem.

Having teammates do tasks xx% faster is zero-sum and chases a local maximum. Faster horses instead of cars. The returns diminish. Efficiency has us question how much work we can get done when we should really be asking ourselves what work should exist at all.

How we do work drives how much we can do.

AI works for consumer use cases like product comparisons because the context is self-contained. The problems we run into as individuals can easily be written into a prompt. Work is considerably choppier. At work, systems fragment context across dozens of tools, owners, and handoffs. The result feels less like a flow and more like a relay race, where the baton gets dropped between every leg.

You can be efficient at the wrong thing. Being efficient is like being fast: there is no direction.

Autonomy Is Leverage

Instead of efficiency, you should seek autonomy. Autonomy means work progresses without constant human orchestration. Speed, quality, and cost are downstream effects, not goals in themselves.

The problem with keeping humans permanently in the loop is not that humans are slow. It is that every handoff forces a context switch, and context switching is the most expensive thing in modern work. It is the reason your best people spend their days routing information instead of using their experience to identify new opportunities.

Autonomy frees attention. And attention is your scarcest resource. When systems carry the orchestration, people stop being infrastructure and start being innovators. They teach their tools instead of serving them. They pursue the frontier instead of maintaining the status quo.

Autonomy unlocks the attention span to innovate. Humans find leverage when they can teach their tools. They pursue the frontier and grow the pie.

Autonomy compounds in a way efficiency never can. Each cycle teaches the system something the next cycle uses. Efficiency has a ceiling. Autonomy has a learning curve.

The Self-Driving Organization

Self-driving cars are measured by level of autonomy, not features. There are five levels, each defined by which human bottleneck it removes. At level 1, you are the driver. No matter how many times you’ve been down a road, your hands remain on the wheel and foot engaging the gas and brakes.

Today, most companies are entirely human driven. We’re scribes, information routers, overcommunicators. Only one week a quarter is spent on strategy, the rest on execution.  The self-driving organization inverts this. The system gathers, synthesizes, and acts. The human reviews, redirects, and teaches. New situations become opportunities instead of overhead. The more the company has seen, the more its institutional knowledge compounds. Autonomy allows people to pursue art over administration.

Why Going Full Stack Matters

The biggest gains from autonomy are reimagined roles. Efficiency takes the org chart as a given and tries to speed up what’s already there. Autonomy asks whether the roles themselves are right.

Software engineering already proved this. Frontend and backend used to be separate disciplines with entirely separate stacks. Full-stack engineering has collapsed that divide. When individuals maintain entire context, they make faster and better decisions and own flows end to end.

At Shepherd, we applied the same playbook to underwriting. We don’t have underwriting assistants. We have full-stack underwriters who handle multiple lines of business, production, and technical underwriting. One person wearing many hats, because our platform carries what used to require dedicated support roles.

Collapsing responsibility increases speed and quality.

This is what going full stack actually means. Owning the workflow so that roles can grow instead of multiply. The biggest rocks become movable when you stop working around your org chart and start reshaping it.

Shepherd as an Example

Shepherd is a vertical services company insuring some of the largest construction projects in the United States. We help everything from data centers to hospitals to schools get built. Pricing a billion-dollar project is a statistics and game theoretical problem. You never have perfect information, and the information you do get is buried in deeply unstructured PDFs and spreadsheets. Insurance is a stress test for autonomy: messy data, intricate regulations, judgment-heavy decisions, and an information supply chain that still runs on email.

At incumbents, each risk is processed through dozens of stages and multiple global teams. Data is processed manually and teams re-review the same information, each adding an additional opinion and slowing things down. Receiving a quote from them can take weeks.

Pursuing efficiency makes each team faster. The autonomous version asks why there are three teams at all. We’ve encoded the industries data structuring, referral triggers, and compliance rules into our own Insurance OS that unlocks the autonomy of full-stack underwriters. Autonomous underwriters have a wholistic view of each risk and deliver lower loss ratios and faster turnarounds. They can focus on the programs that make us a value-add partner and not just another piece of paper.

Our people focus on what makes us different because our platform automated what makes us the same.

Finding Your Level 5

To find your level 5 autonomy, start by setting a destination. For Shepherd, it’s FAU: Fully Autonomous Underwriting. Underwriters become portfolio managers, picking regions and owning their own risk strategies. The platform executes and orchestrates their mandate. Your version will look different, but the exercise is the same: define what “fully autonomous” looks like for your domain, then work backward.

When developing your roadmap, ask yourself:

• Is company context easy to retrieve?
• Are authority scopes and decision makers clearly defined?
• Is there a mechanism to institutionalize new knowledge?
• Can we streamline exception handling?

The biggest rocks are organizational. Collapsing a handoff. Merging two roles. Creating systems that carry what used to require a dedicated person. Unlocking each level requires an evolution of teams and tools. Doing so creates unstoppable momentum.

The Cost of Waiting

None of this is easy. It requires real organizational change, not just new software. But waiting only increases complexity. Autonomy compounds unevenly, which means early movers build learning curves that others cannot shortcut.

The goal is not to make space for AI. The goal is to make space for people to do what only people can do.

If this resonates, feel free to copy it yourself. We hope that along the way, you find your footing and grow your teams, tools, and AI together.

‍