To the Google AI Product Team,
As a dedicated subscriber since the early Bard era, I am writing to formally raise concerns regarding the recent transition to compute-based usage limits within the Google AI subscription tiers.
When I committed to a paid subscription, the value proposition was built on predictable, quantifiable access. Before the recent update, the Pro tier provided a fixed daily prompt allowance that reliably sustained a full workday of development. The shift to an opaque backend compute consumption model, governed by a rolling 5-hour refresh window, has fundamentally altered that agreement.
For developers relying on Gemini for software architecture and technical problem-solving, this creates severe workflow instability. A single debugging session can now unexpectedly exhaust a quota in minutes. For instance, during a recent session diagnosing a multi-layered Python and Kivy UI state issue with an underlying SQLite schema, one hour of iterative stack trace analysis completely depleted my 5-hour limit, bringing my workflow to an unpredictable halt.
This issue is compounded by the structural reality of software development, which inherently requires maintaining large context windows to process syntax, logic, and interdependencies. As debugging cycles progress, iterative context bloat is unavoidable. Penalizing this necessary context retention as a high-compute liability renders advanced development workflows practically unsustainable.
I recognize that advanced AI capabilities require substantial infrastructure, and that context is understood. However, deterministic text and code workloads are computationally distinct from heavy media generation. Drawing both from the same opaque pool makes it impossible to plan, budget, or maintain consistent use of the platform at a professional level.
I respectfully urge the product team to implement the following structural solutions:
- Restore a transparent, quantifiable access tier designed for developer workflows, utilizing explicit token counts or predictable context allowances.
- Implement compute segmentation that strictly separates standard text and code processing from intensive media generation.
- Expose a real-time usage meter, or provide a meaningful telemetry breakdown of how quota is consumed during a session.
Long-term users represent a sustained vote of confidence in this platform’s technical direction. The current implementation risks converting that confidence into attrition. I submit this feedback as a constructive signal to restore predictability for the developer base that helped build your early adoption.
Respectfully,
Hendrick “avid” Hulleman