The 'AI psychosis': the real risk your SMB is running
Aaron Levie, founder of Box, points to a structural problem we see regularly in our consulting work: the decision-makers deploying AI rarely understand how work actually gets done in the field.
ClickUp illustrates this perfectly. The company cut 22% of its workforce betting on AI agents to replace human work. The result? Likely a cascade of operational problems not yet public, but predictable: the agents miss implicit context, edge cases, decisions that require domain expertise and judgment.
This ‘AI psychosis’ is the illusion that because an AI generates a coherent response, it can substitute for domain expertise. It’s especially dangerous for SMBs because:
- Processes are often implicit (“that’s just how we’ve always done it”)
- An AI error directly impacts profitability
- There’s no margin to absorb failures
Levie poses a simple but radical question: who in your leadership actually does the work you’re considering automating? If the answer is “no one in management,” you’re already infected by this psychosis.
What this means for your business
What this means for your SMB
Before deploying an AI agent or automating a process, force yourself to answer this: who in my management understands this process in detail?
If you can’t name them, the AI will fail. Not from lack of power, but from lack of a realistic specification.
The healthy approach for an SMB:
- Involve those who do the work in defining your AI use cases
- Test in parallel for a minimum of 3-6 months (no day-one switchover)
- Keep humans in the loop for decisions that truly matter
AI agents work better as assistants that boost productivity than as replacements that destroy it.
In brief
AI agents vs design: Adobe tests a middle-ground approach
Adobe is launching a conversational AI agent for graphic design that works like an “average intern” rather than fully automating the task. The benefit: keeps the designer in the creative loop. For SMBs with occasional design needs, this is more useful than a “set and forget” image generator.
Cognition (Devin): the CEO of a leading coding AI clears the air
Scott Wu, creator of Devin (an AI agent for coding), explicitly states that code agents shouldn’t replace developers but assist them. While some startups are cutting 20% of dev staff thanks to AI, he acknowledges the limitations. An important signal for SMBs looking to recruit or retain tech talent.
Prompt engineering: two different jobs with one confusing name
The term ‘prompt engineering’ conflates two distinct activities: fine-tuning prompts for an existing LLM (low-level) versus designing enterprise AI workflows (high-level). This confusion creates false expectations when hiring or building an internal AI strategy.
Groq raises $650M and pivots strategy: inference becomes the real business
After Nvidia’s massive acqui-hire ($20B), Groq is changing course: less hardware, more inference optimization. For SMBs, this means AI agent execution costs will likely drop in 12-18 months. No rush to deploy at a loss today.
The Google Gemini Spark paradox: useful but no clear strategy
Google’s 24/7 assistant automates everyday tasks (summaries, planning) but Google doesn’t know how to position it. Symptomatic of a confused market. SMBs should be cautious about integrating a Google product without a real product roadmap.
Get The AI Brief in your inbox
3x per week, the essentials of AI decoded for business leaders.