As of late December, frontier model quality and AI-assisted coding have passed an inflection point. A watershed moment comparable to the dot-com era.
This means:
- “Coding” as a job will largely disappear.
- Technical execution will become very cheap and extremely fast (it already is for many types of tasks).
- Product design will still matter, but because everything can be copied within hours (instead of months or years), it will no longer be a sustainable MOAT.
What it means for currently running businesses:
- Competition will intensify; new companies will compete on price while providing “good enough” services and products.
- A lot of niche software will be built in-house rather than bought off the shelf.
- Valuations of many large organizations will decline over time.
- Companies with extreme lock-in (regulated markets, deep integration, bureaucracy) may persist for a reasonable timeframe.
- Tough times ahead for businesses that are not already operating at the edge of this shift, especially hardware-backed platforms.
What it means for industries:
- Operations-oriented companies will retain their advantages (Uber, Grubhub, etc.), and may even improve.
- Social/network-effect companies will retain their moats.
- Regulation-dependent companies may survive for a few years (3–5) but will struggle over a 10-year horizon, except for a few that solve the AI–regulation challenge.
- Private AI solutions will be popular, but when building solutions becomes effectively free, the role of software companies will change. We may see growth in consumer devices like Omi (AI wearable) and private on-prem inference systems (think NAS-like devices for private AI).
- Hardware + software combinations will thrive; hardware will become more important, will it be back to its glory days?
- Application PaaS and fast “code to production” platforms will grow.
- QA, security, and UAT will be important during the transition but may decline long term as AI improves these functions. Development platforms will increasingly support AI, making secure code more common and automating much of the remaining attack surface.
What it means for AI startups:
- Aside from core AI and frontier model developers, many specialized AI companies will be challenged by powerful general models. Why use a niche solution if a frontier model can produce high-quality documents, perform RAG on your drive, and write code at a very high quality already.
- Fine-tuning and related techniques are becoming commoditized; as long as you have data, customization is easier.
- Startups without proprietary or large data sets will face an uphill battle.
- The IP M&A market will change. Why spend millions on acquisitions when similar capability can be built in-house faster than negotiating a deal?
What it means for org structures:
- QA, development, and creative teams will shrink where AI can replace routine work. Organizations will favor smaller, more agile teams because communication overhead grows with headcount. Brooks’s Law will be very important for orgs’ agility. Everyone will try to avoid and explosion of communication channels as much as they can.
- High-level roles such as software architects, creative directors, product designers will be more important. Strong social and leadership skills will be crucial.
- Marketing and sales will become more critical than ever!
- Director-level positions may reduce middle management layers. Many traditional roles (data engineers, some product tasks) will be automated or augmented by AI. Why hire all these people if AI can do it well enough?
- Domain experts will remain crucial, especially during the transition period.