'Human in the Loop' is a Fantasy. Worse, it Can Be Dangerous to Your Business.

Perspective — Sept 16, 2026

By Piers Fawkes · Founder, Fodda & PSFK · 5 min read

"Human in the loop" is a fantasy. Worse, it can be dangerous to your business. It seems to me that companies adopt this workflow because of a philosophical idea, not a business decision. And so, the math doesn't work.

Every conference panel, boardroom pitch, and vendor whitepaper over the past two years has repeated the same reassuring mantra: "Don't worry—we'll keep a human in the loop."

It sounds responsible. It satisfies the risk committee. It comforts executives anxious about autonomous systems going rogue. There is just one fatal problem: the math does not work.

The Math Doesn't Work: Why Post-Hoc Review Fails

Human In The Loop (HITL) processes rely on a simple premise: an AI system does the work, and a human sits at the other end to review and approve what was created before it ships.

How can that realistically happen in production? AI agents produce far more output than any human team can review. A single agent swarm running overnight can evaluate 500 consumer trend signals, generate 50 brand positioning drafts, or audit 2,000 regulatory compliance filings before your first team member arrives at 9 AM.

You are not going to hire thousands of people to check the work of dozens of agents. That is not a business strategy. That is a mission-critical bottleneck, built by well-meaning people insisting on a process they read about in a newsletter or heard on a podcast.

When organizations attempt to enforce manual post-hoc review on high-throughput agent swarms, one of two dangerous failures inevitably occurs:

• Failure Mode 1: The Choke Point — Your highest-paid strategists, lawyers, and directors spend their days triaging endless review inboxes. The entire velocity advantage of AI collapses to human reading speed.
• Failure Mode 2: Phantom QA — Faced with 200 drafts per day, reviewers experience cognitive exhaustion. They skim, assume it looks plausible, and hit "Approve." You now have zero real quality control masquerading as safety.

The Antidote: Humanity In The Machine

So how do you actually maintain quality and validate AI output at scale?

After 12 months building this at Fodda—and more than 22 years structuring commercial intelligence at PSFK—here is what I believe: You don't need a human in the loop. You need Humanity In The Machine.

For PSFK's Fodda, we built a proprietary system to digitize real experts: people with genuine knowledge, deep industry context, and hard-earned experience, inside or outside your organization.

It works through structured interviews, mental model profiling, and continuous real-world updates, turning their judgment into something durable and queryable. We call the result a Human Agent.

Real Specialists, Digitized and Queryable

A Human Agent is not a generic prompt persona. It is not an LLM asked to "act like a retail strategist." It is a deterministic, structured knowledge graph and reasoning layer derived directly from real practitioners like Jeremy Bergstein, Chris Johns, Alison Cossette, Jeff Squires, Sledge Smith, and Thiago Bersou.

When you digitize specialists, you capture tacit rules of thumb that never appear in standard web scrapes: nuanced taste and instinct, negative constraints (knowing what not to do based on past failures), and deterministic evaluation rubrics.

How a Human Agent Operates Inside the Process

Here is how it actually works in enterprise architectures: A Human Agent plugs directly into your AI workflows—not as a separate review step after the fact, but as a live layer inside the process.

When an agent produces output, the Human Agent checks it against real expert judgment before it moves forward, catching what generic AI would miss and giving you a deterministic, quality-controlled result, at the same speed the rest of the system runs at.

That is how you get expertise into the system without becoming the bottleneck.

Scaling Judgment, Not Manual Labor

This does not eliminate the human expert—it liberates them. Instead of wasting 40 hours a week proofreading repetitive agent outputs, your best minds spend their time conducting high-level research, refining their mental models, and feeding fresh perspectives into their digital twin.

Their human judgment is deployed simultaneously across tens of thousands of automated workflows, multiplying their reach while protecting the organization's standards.

To see how we're building humanity in the machine, book time directly with Piers Fawkes at https://cal.com/psfkpsfk/fodda-human-agents.