What's stable, what's slow

What will almost surely still be true in two years#

1. Businesses still need the same things: customers, cash flow, bookings, bookkeeping, a working website. AI changes how, not whether.
2. The gap between what AI can do and what businesses actually use gets wider. Capability improves every few months. Adoption moves at human speed: trust, habits, training, connecting AI to data. That gap is the most stable opportunity there is.
3. Existing things stay around. Millions of WordPress sites won't vanish in 2 years, because moving is painful. They still need updates, security, speed and fixes.
4. Busy owners want results, not tools. A dentist won't learn an AI coding tool. They'll pay someone to make it work, and want a person to call when it breaks.
5. Accountability stays human. When something breaks, customers want someone responsible, not a chatbot.
6. AI builds more stuff that needs fixing. Sites and apps built with AI builders need hosting, security, debugging and upkeep.

Not stable, so don't build on it: any specific AI tool or wrapper, any specific platform format, search traffic as your main channel, selling information/templates/code/content, which AI lab is leading.

In plain words: sell a result to people who already have a problem, and use whatever AI tools are best that month as your cost. Tools can change every 6 months while the business stays the same.

Why change is slow: history#

• Electricity. Electric motors existed from the 1880s. Factory productivity only jumped in the 1920s, about 40 years later (economist Paul David). The motor wasn't the slow part. Factories had to be redesigned around it: layouts, workflows, retrained workers. AI is at that stage now: companies swap tools but haven't redesigned how they work.
• Websites. Squarespace launched in 2003 and Wix in 2006. For ~15 more years, businesses kept paying developers and agencies. Owners didn't want to learn, wanted it custom, and wanted someone to call. AI today is roughly where websites were around 2004.
• Amara's law: we overestimate a technology's effect in the short run and underestimate it in the long run.

Pace layers#

Stewart Brand's idea: some layers of the world change fast, some slowly. Your job is to connect them. Don't sell the fast layer; use it as your cost. Sell into the slow layers.

On language: AI already speaks most languages, so language alone closes fast. What closes slowly is everything around language: local chat apps, local marketplaces, payment methods, invoice rules, local trust.

When the window closes#

• Pure technical integration closes faster. Vendors ship connectors; agents do setups themselves. Don't be "the person who connects X to Y".
• Process, trust, local knowledge and accountability close slowly. Nobody ships "redesign my clinic's front desk around AI and be responsible when it breaks" as a feature.

So the position is: "the person who makes AI actually work for [niche] businesses in [place], and stays responsible for it."

How to use the window#

1. One niche, one place. Narrow makes you findable and lets you learn fast.
2. Sell a result, not a tool. "Your customer messages get answered in 2 minutes", not "I'll install a chatbot".
3. Work inside what they already own. Their site, their data, no migration.
4. Write down every setup. After 5 clients in one niche, what repeats becomes a product; what you learn becomes posts.
5. Remember you sell to late adopters. They buy on proof, peer references, fixed prices and hand-holding, not vision. Your first 3 clients matter most.

Keep upside without betting: the barbell#

Nassim Taleb's barbell:

• ~80–90% of effort on the stable core. It pays the bills and is very unlikely to break in 2 years.
• ~10–20% on small, cheap bets (a new platform's plugins, a packaged AI tool). If one fails, you lose a few weeks. If one hits, the core already has customers to sell it to.