The Golden Age of AI Applications

We’re entering the golden age of AI applications. Three recent developments confirm it.

The Fable retraction shows regulatory risk. Nadella’s thesis shows strategic consensus. Salesforce’s acquisition shows market validation.

First, the US government shut down Fable access¹ & the software ecosystem roared with many responses : Bring it back! Open-source & local models have become essential! Don’t rely on a single model!

Satya Nadella published an AI ecosystem thesis.² He argued that for a healthy ecosystem, the moat can’t be the model. Instead, human expertise & the system around the model (the harness³) must be the moat.

And Salesforce announced the acquisition of Fin, formerly Intercom, for $3.6b.⁴ The founders & management team repositioned the company through the AI upheaval. Fin used open-source models to maximize price/performance.

Building AI applications is hard for different reasons than SaaS. It’s not a lack of engineers, or the challenges of uptime, or the demands of faster releases.

AI applications present three new disciplines to master : picking the right models, developing the hill-climbing loop, & evaluating the performance of the system for each company, all of which answer the question how much intelligence can I squeeze out of my token budget?

Models are tricky. Budgets prevent defaulting everyone to state-of-the-art. The legion of other models each have a personality. Kimi K2.6 is fast & a great creative writer but less precise. Qwen 3.6 27b is a small model with legendary performance, but it’s a bit of a donkey. It stops suddenly in the middle of a toolchain call & requires a good prodding to push on. GLM 5.1 is an excellent coding model, but a plodder.

Loops, the critical problem-definition exercise of this era, are hard to design. Systems design is an entire discipline (see Donella Meadows’ excellent work on it⁵). What is the best way to define a loop so an agentic system improves? This field is novel & challenging because the models & infrastructure move quickly.

Evaluating the performance of model + loop is ongoing labor. Most companies won’t want to staff a team for each workflow software in a company. AI systems are complex, finicky engines.

The nuances of tuning the carburetors & the timing belts of these complex beasts are tasks better assigned to a few vendors to deliver maximum intelligence per dollar⁶ & amortize the costs across a broader population.

The companies that master these three disciplines will own the golden age.