Autonomous enterprise
- cutting through the hype

Theodore Roosevelt famously said, “Speak softly and carry a big stick.” Less bluster, more substance. Back your words with the strength to act.

In today’s conversation about the “autonomous enterprise,” the big stick is strategy and execution: business-case rigor, process design and human-machine partnerships that differentiate technology from enduring competitive advantage.

Traditional automation follows prescribed steps; agentic AI can plan and act toward goals. Autonomy describes how much authority machines have to decide and act without human intervention. But autonomy is not itself a business objective.

The question is what degree of automation creates meaningful business value, translates into competitive advantage and justifies the execution risks. Initiatives must be demonstrably valuable, competitively advantageous and risk-optimized.

Master these disciplines; learn from every turn of your automation journey. That's the difference between leaders who seize the opportunity and those who merely describe it.

Demonstrably valuable

Too often, the demo precedes the business case. Time saved is treated as cash earned, while integration, oversight, errors and exceptions vanish from the math. Sometimes the best answer is to simplify a process - or not automate it. Value extends beyond direct economics: growth, customer loyalty, talent and resilience can matter more than cost reduction.

Consider a hypothetical automotive dealer network deploying a service agent. Five potential sources of value are: recovering missed inquiries as repair appointments; booking after-hours customers immediately; keeping customers informed; freeing advisors for higher-value consultations; and handling call spikes during recalls.

Proof needs to follow each claim: incremental gross profit after costs, faster booking and completed appointments, satisfaction and repeat visits, advisor time reallocated and successful human escalation. The evidence remains illustrative and uneven: stronger for growth and speed, early for customer value and untested for resilience.

These are value hypotheses, not realized benefits.

Competitively advantageous

An attractive functional ROI can be a poor enterprise outcome. A purchasing agent might lower unit prices through larger orders while mismatching supply to demand, reducing flexibility and causing stockouts. While Procurement hits its target, customers pay the price. A narrowly targeted improvement may be transformative. The test is whether it translates into a superior customer proposition or sustainable competitive economics.

Take promotion planning at a large cosmetics brand. Agents accelerate analysis and execution while people set positioning, select trade-offs and approve material decisions.

Before agent deployment, the brand manager chooses from last year’s calendar. Analysts reconcile spreadsheets; sales and finance negotiate through repeated approvals. Operations learn from late, fragmented reporting.

After, the brand lead sets the brand promise, promotion goal, margin floor and privacy rules. Agents model offers, and ensure against cannibalization and stockouts. Humans approve trade-offs, retailer terms and exceptions. Agents publish approved offers, monitor uptake and flag shortages.

The customer gains more relevant offers, better on-shelf availability and faster response to trends. The critical measures are incremental contribution margin, full-price sell-through, time to launch, availability and repeat purchase.

GM’s 1980s robotics effort failed to deliver the advantages of better-designed production systems. C.H. Robinson reports over 80,000 ocean and air-freight quotes for over 3,600 customers, cutting turnaround from days to minutes: company-reported speed, not independent proof of market-share gain. Amazon’s million-robot milestone matters for better speed, reliability and economics, not robot count.

Risk optimized

Plot decision complexity against the consequences of a wrong decision, including severity, propagation and reversibility. The quadrant is an initial hypothesis for human-machine design, not a contest to remove people.

For an industrial pipe-and-fittings manufacturer, classify tasks, not departments. Human approval points, exception rules and auditability follow the risk.

Where complexity and downside are lower, closed-loop automation can acknowledge standard orders and route invoice queries. Mistakes are cheap to correct.

Where complexity rises but downside remains lower, bounded agents evaluate noncritical packaging suppliers or simulate maintenance schedules. Agents explore without committing; scenarios can be rejected before affecting production.

Where complexity is lower but consequences are serious, updating pressure-rating SKU labels or supplier bank details requires controlled execution. Validation is rule-based, but release requires approval: incorrect ratings create safety exposure.

Where both are high, AI supports expert judgment on safety-rated substitutions or scarce-output allocation after a foundry disruption. Technical, regulatory, contractual and customer trade-offs interact; engineers and executives retain final sign-off. 

Ford said in 2026 it had hired 350 experienced engineers over three years to remedy quality shortcomings automated tools missed. Klarna reports AI handled 80% of service chats in 2025, saving approximately $59 million while preserving human support. Both show how economics, quality and human judgment must be designed together, not assumed apart.

Human review changes the economics; customer failures erase local savings; better models can justify new decision rights. Agents can scale a strong operating model or compound a flawed one. Autonomy is a variable to optimize, not a milestone to celebrate. Walk softly through the hype; let business results do the talking