Themata.AI
Themata.AI

Popular tags:

#developer-tools#ai-agents#llms#claude#ai-ethics#code-generation#ai-safety#openai#anthropic#discussion

AI is changing the world. Don't stay behind. Clear summaries, community insight, delivered without the noise. Subscribe to never miss a beat.

© 2026 Themata.AI • All Rights Reserved

Archive

|

Topics

|

Privacy

|

Cookies

|

Contact
inference-apisai-agentsmodel-interoperabilitydeveloper-tools

The session you cannot take with you

The Session You Cannot Take With You | EARENDIL

earendil.com

July 31, 2026

14 min read

🔥🔥🔥🔥🔥

67/100

Summary

Inference APIs allow users to send input and receive output, facilitating conversation archiving and replaying. However, challenges such as prompt cache storage on external GPUs, differing tokenization between models, and intentional sampling unreproducibility complicate session portability.

Key Takeaways

  • Inference APIs are increasingly returning non-portable session data, which limits user ownership and access to their AI interactions.
  • Features like encrypted blobs, hidden instructions, and provider-bound state prevent users from fully understanding or continuing their sessions with different models.
  • A truly portable session should allow users to export, inspect, replay, audit, and delete their session data independently of the provider.
  • The marketing of encryption features often misleads users, as the data is typically only accessible to the provider, not the user.
Read original article

Community Sentiment

Mixed

Positives

  • This article gives a very good overview of a problem that most users of AI rarely evaluate, highlighting important issues with provider lock-in and user experience.
  • I'm optimistic that dark patterns will eventually be replaced by more respectful practices in AI, as the market matures and user awareness grows.
  • Many commenters are finding innovative ways to maintain context across sessions, suggesting a creative push towards better user experience in AI interactions.

Concerns

  • If switching models is painful, providers can create vendor lock-in, enshittify the user experience, and drive up costs — this is a huge concern for future competition.
  • The coupling of powerful extensions with inference APIs creates unnecessary barriers, limiting user freedom and innovation.
  • Some models are overly verbose, resulting in poor signal-to-noise ratios that frustrate users and hinder productivity.

Related Articles

[PSA] Anthropic's Method to Losing Goodwill in a Few Easy Steps

Anthropic's Method to Losing Goodwill in a Few Easy Steps

Jul 6, 2026

Who’s Afraid of Chinese Models?

Who's Afraid of Chinese Models?

Jul 20, 2026

The Agent Harness Belongs Outside the Sandbox

The agent harness belongs outside the sandbox

May 2, 2026

Our position on open-weights models

Our position on open-weights models

Jul 27, 2026

Local AI Needs to be the Norm

Local AI needs to be the norm

May 10, 2026