The 80% Is Commoditized

My X feed is full of people shipping AI tools every day.

Open source projects trending on GitHub overnight. ClaudeHUD - a plugin that shows you what your coding agent is doing - went from zero to trending in days. Garry Tan open-sourced gstack - his Claude Code workflow - and it hit 20K stars in a week, trending on Product Hunt and sparking massive debate. OpenClaw hit 210K stars in two months. Solo builders replacing enterprise SaaS products with $200/month of API calls.

It feels like everyone is building.

But I have to check myself: my feed is a bubble.

The people I follow on X are the top 0.1% of AI adopters. They’re shipping agents, open-sourcing tools, and building entire businesses with 13 AI agents. They make it look like the whole world has moved to AI-native workflows.

The reality is very different.

Most companies are still using AI as a chatbot. A smarter search engine. A writing assistant that drafts emails. The gap between what AI power users are doing and what the average knowledge worker is doing is enormous - and growing.

That gap is the real opportunity.

The tools trending on GitHub this week solve problems for builders. Developer productivity, agent orchestration, coding workflows. That’s a real market - but it’s a small one. 4.3 million AI repositories on GitHub, thousands of agent tools, and they’re all competing for the same audience of developers and early adopters.

Meanwhile, the finance team at a mid-size company is still copy-pasting numbers between spreadsheets. The sales team is still manually qualifying leads. The operations team is still running processes that haven’t changed in five years.

The bottleneck has shifted. Building is cheap now. Anyone with an API key can get to 80% of a working solution in a weekend. For a lot of use cases, that 80% is genuinely good enough. Content drafts, internal dashboards, workflow automation, prototyping - AI handles these well because the cost of being slightly wrong is low.

The hard part is the other 20%.

That’s where decision quality matters. Financial calculations that need to be exact. Compliance workflows where “mostly right” isn’t acceptable. Sales intelligence where a wrong recommendation burns a relationship. Any use case where the output directly drives a high-stakes decision.

This is where everything breaks. The context window doesn’t hold enough of the problem. The agent forgets what it learned yesterday because there’s no persistent memory. The output looks right but is subtly wrong because the system has no understanding of the domain constraints, the edge cases, or the business rules that can’t be inferred from a prompt.

The 80% is commoditized. The next generation of agentic software will differentiate on the last 20% - memory management, context engineering, and precision in workflows where decision quality is everything. Not model capability. Not UI. How well the system remembers, reasons over context, and handles the details that actually matter.

The distribution problem compounds this. The real constraint isn’t marketing. It’s reaching the 99% who don’t know these tools exist, don’t know what’s possible, and wouldn’t know where to start.

But let’s be honest about what drives distribution too. Gstack didn’t just trend because it was useful. It trended because the CEO of Y Combinator built it. ClaudeHUD trended because it solved an obvious pain point every Claude Code user had. OpenClaw trended because setup took minutes instead of hours.

Three different distribution engines: personal brand, pain point precision, and frictionless onboarding. All three work. None of them are “run ads and hope.”

Product-led growth has always mattered. In the AI era, it’s the only thing that matters. The product has to be so obvious, so frictionless, and so clearly valuable that it spreads without anyone explaining what an LLM is.

The biggest companies of the next decade won’t be built by the best AI engineers. They’ll be built by the people who close the last 20% - with real memory, real context, real precision - and distribute it to the people who need it most.

The 80% is commoditized. The last 20% is where the next decade gets built.