Composable AI: Build Prod, Not God

TRANSMISSION RECEIVED · PLANET COMPOSE-1 · COORDS [-0.55, 0.12]

A new AI lab, TypeSafe, is operating in stealth, and its manifesto opens with a bracing thesis: “The only evidence AGI isn’t already here is the embarrassing lack of impact modern AI has had.” After trillions of dollars, most software still is not meaningfully intelligent. The bottleneck, it argues, is not raw intelligence — it is that today’s intelligence is hard to build on.

Intelligence is everywhere and in nothing

The framing is the “horseless carriage”: early cars kept the high seats, buggy springs, and even the whip socket while swapping the horse for a motor — a new technology forced into the shape of what it replaced. Today’s AI is trained to be a helpful, articulate, pleasant assistant. That goal makes sense if a human is always on the other end. The predictable consequence is AI that needs humans in the loop instead of running quietly in the background.

Software has never worked that way. Even the most complex system is built from simple, auditable logic and layered abstractions. TypeSafe wants AI to work the same way — as a primitive any programmer can invoke for semantic judgment and decisions, while code still does what it is best at: exact computation. Computers branch on bits today; imagine if they could also branch on intent, common sense, and understanding. That is the neuro-symbolic dream, occasionally snarkily compressed to “smart if-statements.”

Intelligence is like databases before SQL

The people who built databases did not imagine Google, and the people who built internet protocols did not envision Stripe. They made lower-level capabilities so dependable they could run unattended and be layered on top of — an unplanned Cambrian explosion no one designed top-down. Intelligence today is like databases before SQL: powerful, but every use is bespoke. Once a smart decision becomes as dependable and invokable as a database query, builders will stack them the same way.

Safety is the precondition for composability. You let a component run unattended only if it is reliable, and you build on top of it only if it is trustworthy — it takes trust to bury a dependency five layers deep in a system. So TypeSafe’s path is to ship machine-native composable AI with the highest intelligence-per-dollar, make it reliable enough to transform the economy through real automation, then expose higher-level intelligence abstractions stable enough to compose and layer upon. Their slogan: “We’re building prod, not God.” Their measure of success: global TFP growth reaching 3% and holding for a decade — unprecedented in economic history.

Why marketers should care

Most marketing software that uses AI still treats the model as a chatty assistant a human nudges, checks, and rescues. The composable agenda reframes it: what if the intelligence behind personalization, asset tagging, copy variants, and campaign decisions were as dependable and callable as a database lookup — auditable, testable, running in the background? Precisely because it is silent and reliable, teams could layer more of it into production content pipelines instead of babysitting prompts. The horseless-carriage warning applies to agencies too: bolting a chatbot onto a website and calling it AI keeps the old shape; composable AI changes the shape of the workflow.

How to use it

• Look for where you currently keep a human in the loop for judgment calls a model could own if it were reliable enough; those are composability candidates, not a risk too far.
• Favor AI integrations that expose auditable, testable calls over opaque assistant chatbots you cannot constrain.
• When evaluating vendors, ask not “how clever is the model” but “how safely can I stack on top of it.”

// END OF LOG