Definition
The Forklift and the Coffee Cup is Greg Kihlström‘s image for the over-provisioned-mechanism anti-pattern in the book Stop Saying “AI”. A forklift is the wrong tool for carrying a coffee cup across a room, even though it’s the more powerful and more impressive machine. Using it is slower, costlier, and riskier than just picking the cup up.
In AI terms, it names the recurring mistake of putting a heavy mechanism — a generative model, or an agentic loop — under a task that a rule or a predictive model would clear deterministically and for a fraction of the cost. The capability gets delivered, but at the wrong price, with more variance and a harder-to-audit failure surface than the job ever needed.
The forklift is the counter-image to the framework’s operating rule, the lightest-mechanism discipline. Where the discipline says use the lightest mechanism that clears the task’s accuracy bar, the forklift is what you get when a team ignores the bar and reaches for the heaviest mechanism available.
How It Relates to Marketing
Marketing stacks are full of forklifts, because the pressure to “use AI” often lands as pressure to use the most advanced-sounding AI. A templated, rules-friendly task — routing tickets by keyword, sending a triggered email on a known condition, scoring a lead on structured fields — gets rebuilt on a generative or agentic mechanism because that’s what the roadmap called for. The output looks the same to the customer. The bill and the risk don’t.
Naming the pattern gives a marketing leader permission to ask the unfashionable question in a vendor demo or a roadmap review: does this job actually need the heavy machine, or would a lighter mechanism carry the cup? Often the honest answer trims cost and variance at once.
Why Teams Reach for the Forklift
The pull toward the forklift is mostly linguistic, and it traces straight back to the two collisions. The words “generative” and “agentic” name both a mechanism and a pillar, so when a team hears that a capability involves generation or agents, it reaches for the generative or agentic mechanism by reflex — even when the pillar could be served by something far lighter. The impressive vocabulary does the deciding, and the accuracy bar never gets asked about.
A forklift really is the right tool sometimes. A genuinely open-ended, unstructured, novel task can defeat every lighter mechanism, and there the heavy machine earns its keep. The anti-pattern isn’t using a forklift. It’s using one for a coffee cup.
How to Spot It
- Find the heavy mechanisms. List every capability running on a generative or agentic mechanism.
- Describe the task plainly. Is the input structured and predictable, or open-ended and novel?
- Test the lighter option. Ask whether a rule or a predictive model would clear the accuracy bar for this specific task.
- Price the gap. Compare the compute and verification cost of the heavy mechanism against the lighter one that would still clear the bar.
- Flag the forklifts. Any capability where a light mechanism clears the bar but a heavy one is running is a forklift carrying a coffee cup.
Comparison to Similar Concepts
| Concept | Names | Relationship |
|---|---|---|
| The Forklift and the Coffee Cup | Over-provisioned mechanism | The anti-pattern the lightest-mechanism discipline prevents |
| Over-engineering | Building more than the problem needs | The general case; the forklift is its AI-mechanism form |
| Gold-plating | Adding unrequested capability | Related waste, usually about features rather than mechanism |
| Resume-driven development | Choosing tech to look advanced | A common cause of forklift choices |
The forklift is specifically about mechanism, not features. It isn’t building too much product. It’s building the right product on a machine heavier than the job requires.
Best Practices
- Let the task pick the mechanism, not the vocabulary. If “generative” or “agentic” is doing the deciding, stop and ask about the bar.
- Default to the left. Start at rules and move right only as the task’s openness forces you to.
- Audit for forklifts on a schedule. Over-provisioning accumulates quietly; a periodic sweep catches it.
- Remember the reverse error exists. Under-building an open task is also a failure. The goal is fit, not minimalism for its own sake.
Future Trends
- Cheaper heavy mechanisms hide the waste. As generative costs fall, forklifts get easier to overlook, even as their verification burden persists.
- Cost scrutiny catches up. As AI line items grow large enough to notice, finance-led reviews will start surfacing forklifts that roadmap reviews missed.
FAQs
1. What does the forklift stand for? A heavy AI mechanism — generative or agentic — used for a task a lighter mechanism would clear. It’s powerful, impressive, and wrong for the job.
2. Why do teams do it? Usually because the words “generative” and “agentic” name both a mechanism and a pillar, so teams reach for the heavy mechanism by reflex when they hear those words.
3. Is using a forklift always wrong? No. For a genuinely open-ended, novel task, the heavy mechanism is correct. The mistake is using it for a simple, structured one.
4. How is this different from over-engineering in general? It’s the specific mechanism form of over-engineering: right product, wrong machine.
5. What’s the fix? The lightest-mechanism discipline — use the lightest mechanism that clears the task’s accuracy bar.
Related Terms
- The Lightest-Mechanism Discipline
- The Mechanism Axis
- The Two Collisions
- The Pillar × Mechanism Grid
- Cost-of-Wrong
- The AI Capability Framework (Master Stack)
- Agentic AI
- Generative AI
Sources
- Kihlström, Greg. Stop Saying “AI”. https://amzn.to/4wilWcA
- Greg Kihlström — official site. https://www.gregkihlstrom.com