AI Is Not a Football Team: Why I’m Still Trying to Figure This Out

By Bogomil Shopov – Бого – June 1, 2026

---

The metaphor that keeps coming back

When headlines describe a language model as “scoring a goal” or a reinforcement‑learning agent as “making a perfect pass,” the image that sticks is a football match. The metaphor feels convenient: it gives a complex system a familiar narrative, a set of roles (striker, defender, coach), and an implied scoreboard of success.

But the more we repeat the comparison, the more it shapes the questions we ask. We start treating AI like a player who can be praised for a brilliant dribble or blamed for a missed tackle. That framing subtly shifts responsibility from the people who design, train, and deploy the system to an imagined autonomous athlete.

Where the analogy breaks down

1. No free will, no intent – A footballer decides to shoot because of personal ambition, fatigue, or tactical insight. An AI model produces an output based on statistical patterns in its training data. There is no internal desire, no “will to win.”
2. Training vs. practice – Athletes improve through deliberate practice, feedback loops, and physical conditioning. AI improves through gradient descent on a loss function, a process that is opaque and can be nudged by data selection rather than conscious effort.
3. Team dynamics – A football team has a hierarchy, a coach, a shared strategy, and a culture that evolves over seasons. AI systems are often assembled from disparate modules (LLMs, vision encoders, retrieval layers) that never interacted before deployment. Their “teamwork” is a product of engineering glue code, not a shared purpose.
4. Scoring metrics – In sport, the goal line is clear: more points win. In AI, success is measured by a patchwork of benchmarks, user satisfaction scores, and business KPIs, which can conflict or be gamed.

These mismatches matter because they affect how we allocate blame when things go wrong. If an autonomous vehicle “misses a tackle,” do we blame the “player” (the model) or the “coach” (the data engineers, the safety team, the regulator)? The football metaphor nudges us toward the former, which can obscure systemic accountability.

The real stakes for the future of work

The metaphor also colors the debate around AI‑augmented labor. Some pundits argue that workers will become “assistants” to AI “captains,” implying a clear hierarchy where humans follow algorithmic direction. Others claim AI will be a “partner” that shares the field equally.

Both extremes ignore the fact that most AI tools are assistive extensions – they surface information, automate repetitive steps, or generate drafts that humans still need to edit. The value they add is highly context‑dependent and often requires domain expertise to interpret correctly. Treating the relationship as a simple coach‑player dynamic oversimplifies the negotiation of trust, control, and compensation that will define future workplaces.

A more useful framing

Instead of a football team, consider AI as a toolbox that a human craftsman reaches into. Each tool has a purpose, a limitation, and a maintenance schedule. The craftsman decides which tool to use, when to switch tools, and how to combine them.

• Agency stays with the human – The decision‑maker remains the person who selects the model, curates the data, and sets the operating parameters.
• Responsibility is distributed – Accountability is shared among data curators, model trainers, product managers, and end users, rather than being projected onto an imagined autonomous player.
• Performance is measured by workflow outcomes – Success is judged by how well the tool integrates into a broader process, not by a single metric like “goals scored.”

What this means for developers and policymakers

1. Documentation over hype – Teams should publish clear model cards that explain training data provenance, known biases, and failure modes, rather than glossy “player stats.”
2. Regulatory focus on the supply chain – Oversight bodies need to trace responsibility through the data pipeline, model versioning, and deployment environment, not just the final output.
3. Education that demystifies the math – Teaching developers the fundamentals of gradient descent, loss functions, and statistical inference reduces reliance on metaphor‑driven intuition.
4. Design for human‑in‑the‑loop – Interfaces that surface uncertainty, allow easy correction, and keep the human in control prevent the illusion of an autonomous “star player.”

Closing thoughts

I keep returning to the football metaphor because it’s easy to grasp, but the more I examine it, the more I see the cracks. AI is not a team of players with agency; it is a collection of statistical instruments that amplify human intent—when used responsibly.

The challenge is not to discard all analogies, but to choose ones that preserve clarity about who is deciding, who is accountable, and what the real performance metrics are. Until we get that right, the conversation will keep circling back to the same tired playbook.

---

Follow Bogomil on Twitter for more musings on open source, security, and the messy reality of AI.