In this week’s Big Ideas, I want to highlight two essays that have been getting attention online this week.
First, “Something Big is Happening” by Matt Shumer, who argues this is a near-term, broad disruption to knowledge work, not a distant possibility, and cites AI’s growing ability to write, test, and iterate on software as the leading indicator that similar capability jumps will rapidly spread to law, finance, healthcare, writing, and customer service.
Key takeaway: people and organizations should stop judging AI based on older or free-tier experiences and instead start using current, top-tier tools directly in real workflows, because early adoption and adaptability are positioned as the main advantages during the transition.
Practical recommendations include paying for leading models, applying them to high-value tasks (not just Q&A), building “learn fast” habits, reducing financial fragility, and leaning into work that is harder to replace (relationships, accountability, regulated sign-off), while recognizing these are likely temporary buffers rather than permanent protection.
In a recent article titled “The AI Vampire,” Steve Yegge discusses the growing issue of burnout caused by the increasing reliance on artificial intelligence in the workplace. He highlights a troubling trend where employees, encouraged to maximize productivity through AI, are experiencing exhaustion and a lack of value capture for their efforts, ultimately leading to burnout. Yegge argues that while AI tools can significantly boost productivity, the benefits often do not translate to workers, leaving them drained and unrecognized. He notes that this phenomenon is particularly prevalent in tech environments where companies push employees to adopt AI without properly balancing workload and well-being. According to Yegge, the solution lies in redefining the workday to focus on quality over quantity, advocating for shorter working hours to combat the pervasive fatigue associated with intensive AI use.
Why do we care?
Shumer’s selling the promise, Yegge’s warning about the consequences, and MSPs are about to get caught in the middle.
Shumer claims GPT-5.3 Codex can autonomously write software with “judgment and taste”—he describes a task, walks away for four hours, and it’s done. His evidence is personal anecdote, not reproducible data. He’s an AI startup founder, so his incentive is driving adoption.
Anecdotes about autonomous coding don’t automatically translate to MSP reality: password resets, access requests, endpoint triage, and documentation have different risk profiles and success criteria than greenfield software tasks. But his directional claim is correct: top-tier AI tools can now handle junior-level work—ticket triage, scripting, documentation. He cites METR trend data as supportive, but even the script notes it hasn’t tested the latest models—so treat the curve as suggestive, not predictive for your service desk.
Yegge’s counter: 10x productivity gains are real, but employers capture the value. Workers use AI to do 3x the work for the same pay, leading to burnout. His solution—3-4 hour workdays—is unenforceable without collective bargaining, which MSPs don’t have. But his $/hr framework is the right lens: if you increase output 2x via AI without renegotiating contracts, your effective hourly rate declines.
The play: Use AI to reduce labor costs, not increase output. Pilot it internally first—if you can’t demonstrate 30%+ efficiency gains on your own workflows, don’t sell it to clients. Renegotiate contracts with AI-enhanced SLA tiers priced as a premium tier based on measured outcomes—for example, a 10–20% uplift if you can document reduced time-to-resolution or lower incident volume. Implement Yegge’s $/hr framework for retention—share a defined portion of verified productivity gains with staff—through comp, time, or training budget—so AI doesn’t become ‘do more with less’ until attrition forces your hand.
Shumer’s “pay for leading models” recommendation creates vendor lock-in. Test open-source alternatives—if they’re 80% as good at 10% the cost, you’ve reduced dependency on OpenAI’s pricing. Productize one AI-native offering by Q4 2026—pick a repeatable task like phishing triage and sell it as a fixed-fee add-on. This creates new revenue instead of just compressing margins on existing services.
Are your contracts designed to capture AI value—or to hand it back at renewal?