AI
The AI Revolution Won’t Happen Overnight
This article was full of insights that don’t break through the media hype around AI. The transformation will be impactful, but will be “slower, messier, and far less lucrative in the short term.”
- AI’s real impact will take much longer than we think
- Measurable efficiency gains remain elusive
- Other “revolutions” like the printing press, electricity, and the internet take time to truly impact the economy
- MIT economist Daron Acemoglu states the challenges for orgs are the costs of disruption, retraining, integration, and computing
- We’ve already picked the low-hanging fruit of digital transformations (automating operations, digitizing info, moving customers online, cloud infrastructure), and each new leap delivers diminishing returns
- US total factor productivity has been sluggish for decades (1974-2024); there may be gains in personal productivity but gains at scale aren’t there
- Value comes from targeted, deliberate integration. Build the right systems, train your team, and figure out how to make AI work for your business.
- We’re being wildly optimistic about enterprise AI adoption
- We’ve seen overheated hype cycles before (personal computers, dot-com bubble, blockchain)
- Planning fallacy — underestimate how long transformation takes
- Optimism bias — believe adoption will be smooth and easy
- Recency bias — believe viral consumer adoption translates to enterprise adoption
- There are many systemic barriers (outdated systems, regulatory roadblocks, risk-averse corporate cultures, AI talent shortages, procurement bottlenecks)
- IBM Watson Health was supposed to “outthink cancer” but was sold for parts in 2022
- The winners won’t be making the boldest claims. They’ll be the ones with the patience to build real, lasting change.
- The market is overestimating the value of AI companies
- Investors think AI companies are high-growth, asset-light software firms (actually capital-intensive, high-cost, and infrastructure heavy)
- Infrastructure demands are staggering, and competition (e.g., open source models) is squeezing margins
- Many industry leaders are making high-stakes decisions based on tools built by companies whose business models may not be sustainable
- Winners will embed AI where it creates durable, economic advantage — speeding up business decision cycles, improves decision quality, or reimagined products. Leadership stamina > speculation.
- The real money isn’t in the models
- Math can’t be patented
- Invention = breakthrough (e.g., transformer architecture, algorithms). Innovation = distribution, margin, market fit.
- AI will be a flashy prototype unless you invest in applications, integration, data infrastructure, workflow redesign, and change management
- The real value isn’t in building AI — it’s in using it. The companies solving complex, industry-specific problems with custom AI will create lasting value. Make AI boring (seamlessly embedded, consistently reliable, and quietly transformative).
- We’re over indexing on startups
- AI isn’t about disruption, it’s about distribution
- Incumbents control enterprise budgets, IT integration, and distribution — you don’t need the best model, just good enough
- AI is capital intensive, infra-heavy, and favors scale; Big Tech already owns the data, compute power, and enterprise relationships
- We’ll likely run out of high quality training data in 1-7 years
- Proprietary, real-time enterprise data is the last true moat
- We’re obsessed with generative AI but it’s not the future
- AI struggles with situational awareness, complex reasoning, and synthesizing multiple types of changing info in real time
- Multimodal and compound AI systems are the next evolution. Invest in data architecture, workflow flexibility, and AI governance that can evolve as the tech does.
Software development
Manage Your Technical Debt LIKE THIS, & Thank Me Later
- Bad code = difficult to reason about, read, and change without having other effects. Use static code analysis tools for this.
- Technical debt = gap between the design as it is and the design you not realize you want to have.
- Strategy 1 — rewrite completely or rewrite parts (think: ship of Theseus)
- Strategy 2 — have a cleanup team. Use hotspot analysis to find things that are high complexity and change frequently.
- Strategy 3 — use gen AI. A recent paper says this is not reliable (37% of changes were helpful).
- Strategy 4 — make refactoring part of your job using proven deterministic tools as part of policy/culture. Tidy first before building the feature.