an internal battle within Apple Intelligence for the control of everything

The unfulfilled promises of Apple Intelligence have left Apple something exposed. The case suggests a company somewhat divided by opposite visions and structural problems that culminated in its greatest public stumbling in years.

The background. The plan to renew Siri completely, presented at the WWDC 2024It turned out to be a fictitious demo that not even the engineers of the Siri team had seen work.

According to a long and detailed report of The Informationthe only real element of the demo was the new interface that appears when invoking Siri. Advanced functions such as contextual postal analysis or integration with applications were practically non -existent.

This break with Apple’s tradition of showing only functional technology has left him in an awkward position: have to delay some of the promised functions Until 2026.

Between bambalins. The investigation of The Information Talk about an internal war between the AI ​​/ML team (artificial intelligence / Machine Learning, ‘automatic learning’) led by John Giannandrearesponsible for Siri since 2018, and the software group of Craig Federight Aka ‘Pelazo’.

This conflict is the result of gathering two opposite philosophies:

1. Giannandrea: Cauto approach, oriented to incremental improvements, focused on privacy.
2. Federight: more agile, pragmatic and results -oriented approach.

Apple engineers sarcastically nicknamed the AI/ml group as “Aimless” (without a goal), referring to generalized frustration with their leadership and lack of ambition.

The crossroads. Siri’s course has been confusing from the beginning, marked by inconsistent decisions. Initially it was proposed to develop two language models for this new version:

• “Mini Mouse” to work at home.
• “Mighty Mouse” to execute in the cloud and oriented to the most complex tasks.

Then, the team pivoted towards a single cloud model, before making other technical changes that engineers did not fit well.

Script turn. In March 2025, Apple changed the chairs, putting Craig Federight and Mike Rockwell (architect of the Vision Pro) at the head of the Siri project. The first important change was to eliminate the restriction of using only internal language models. Now Federight has instructed to “do whatever it takes to build the best functions of AI”, even if that means using third -party models.

This decision contrasts with the previous policy to only use external models as a comparative reference, despite the fact that internal models “did not give up as well as OpenAi technology.”

In perspective. Apple’s stumbling block is the reflection of a major problem: its obsession with privacy and total control has collided with the reality of the development of modern. Rivals like Amazon (with Alexa+), Microsoft (COPILOT) and Samsung have been directly associated with Openai or Anthropic, Apple has tried to build everything internally. The results are not being spectacular.

The new direction implies that Apple has understood that it needs to find a balance between its privacy principles and the need to advance rapidly in AI. It is that or, irremediably, losing relevance.

And now what. Within Apple there is optimism that Federight and Rockwell, known for delivering complex projects on time, can straighten the course.

The next one WWDC 2025 In June it will be key to see if Apple can recover the confidence of developers and users after one of the most delicate episodes in the company’s recent history.

Outstanding image | APPSHUNTER.IO in Unspash

In Xataka | Apple and Trump’s dance is taking shape: threat, panic … and an imminent exemption