AI Native

Agentic AI as built-in is powerful, but ultimately it pollutes and otherwise bloats the most excellent of more deterministic softwares.

The future is better as “AI-ready”—adding surfaces and integration points that allow for extensibility—AI not as the core experience, but an optional enhancement to core software for those that want to exploit their data via the exploitation of the commons.

The funny thing is I think many early adopters of “AI-native” are now less AI-ready than some of their explicitly (or implicitly) anti-AI peers. Devs are moving away from platforms and to more direct LLM access, developing their own harnesses that fit the need and reduce costs. And now we’re starting to see the AI-natives scrambling to keep up. The feature factory is clearly brutal. Not much innovating happening, and users are mostly asking “can you add <x> from Claude” within minutes of release. They don’t even know what would be useful beyond “well, the main players are doing it so it must be good.”

What happened? We were so excited about distributed microservices! Why did most companies land on the idea that agents should come bundled with—nay, required, as the interface to—data? I can think of a few reasons and they’re all monopoly coded.

Tips hat and adjusts monocle.

Obsidian seems to have it right: no discernable AI surfaces at all! But, boy, can you do more with LLMs when your data is offline-first, stored in flat files, and end-to-end encrypted. From an individual’s perspective, I think it’s currently more powerful than any AI-native surface, and they did little but stand on the principles of privacy, ownership, and consent.

Makes you think, no?

Sure, Obsidian doesn’t have the power of multi-player, but leaving this completely open as a “bring-your-own-harness” surely is the future of this technology. How Notion might bridge this gap is definitely the challenge for 2027 and beyond.

TBH, I think software needs to move away from following UI patterns from Anthropic and OpenAI. Software can and should be both AI-free and AI-ready by not being AI-native.

Consent and optionality are key considerations in the best versions of AI futures. There are way too many mixed incentives in built-ins. This model leads to feature bloat, increased costs as metering and token-based pricing comes online, not to mention eventually all AI-natives must seemingly compete with the model providers in some way or form.

Agents should not be “native” to software. They don’t need to be. In fact, they are often worse off for it.