Few AI products compete on quality or cost.
Instead they sell based on vendor lock-in, speed to market, AI FOMO, and how well someone targets you with just the right message. The result: business owners pay hefty markups for thin wrappers on top of the same core models. And resources are diverted away from employees who could upskill to solve the same problems more effectively.
The truth is that 90%+ of the capability comes from the models themselves. A single model update has the potential to shut down whole classes of AI startups overnight — and we've seen this trend repeatedly.
Between the model provider tooling and open source projects, businesses have everything they need to solve many of these same problems themselves. The gap? Knowledge.
For those looking to take control of their AI journey
Document everything. Your SOPs are way more valuable than any agent framework itself, and you can easily copy-paste them between frameworks. Bonus points for writing everything in markdown — or having your agent convert it.
Master your data. Clean it up and capture as much of your processes as possible in the form of data. This gives the models hooks into what actually drives your business.
Stay vendor agnostic where possible. Optimization means combining models from different vendors and open source. Don't lock yourself into a single provider's ecosystem if you can avoid it.
Learn the basics. Prompting, context, embeddings, simple APIs, MCP, skills — these all transfer whether you're building custom or using a vendor product. The fundamentals compound.
Build narrow, testable capabilities. Focus on concrete, incremental value rather than the pipedream of a super agent that does everything. You'll quickly see the value one MVP at a time — and it will build on itself.
The bottom line
Want to win the AI revolution?
Focus on your business and implement a few simple solutions. Be very discerning. Save the dollars, security headaches, and technical debt.