As software engineering continues to shift from humans writing code manually to agent-driven production, a lot has been happening at GitHub since the CEO stepped down in August 2025, particularly with respect to infrastructure pressure, largely due to the fact that agentic coding has longer-running & parallel sessions (and demands more compute).
Fundamentally its architecture is too static to handle the type of loads that are evolving with AI workflows - yet, some analysts assume a more strategic pricing structure will fix these issues.
Since Thomas Dohmke stepped down as CEO, he founded a startup (backed by M12) called Entire.
This new platform introduced an open-source command-line tool called Checkpoints, which completely changes the system concept of version control where just a final cryptographic diff is recorded (with no sequence of events observation).
The race to AGI exhausted model families like Phi-5 on pure internet data, this led to complex methods for training AI - like using logic flow pipelines & cognitive telemetry collection.
Checkpoints don't just manage files, the whole idea is to create a version agent session context - this new orchestration layer will intercept every cognitive process an agent has when an edit is pushed.
Entire showcases how projects will be built using complex pathfinding algorithms that will merge with code generation.
As software is pivoting to agentic logic traces, developers will need to consistently audit their 'intent-checkpoints' and be diligent when it comes to state-space serialization.
May the 4th be with you.