Please support this podcast by checking out our sponsors:
- KrispCall: Agentic Cloud Telephony -
https://try.krispcall.com/tad
- Lindy is your ultimate AI assistant that proactively manages your inbox -
https://try.lindy.ai/tad
- Discover the Future of AI Audio with ElevenLabs -
https://try.elevenlabs.io/tad
Support The Automated Daily directly:
Buy me a coffee: https://buymeacoffee.com/theautomateddaily
Today's topics:
Agents need real cloud environments - Cursor says cloud coding agents live or die by a complete developer-grade environment—missing dependencies and tooling quietly degrade output quality, not just cause errors. Keywords: cloud agents, dev environment, reliability, dependencies.
Durable execution for long tasks - As agents run longer and unattended on dedicated VMs, failures shift from laptops to cloud outages and restarts—pushing systems like Temporal for durable execution, retries, and consistent streaming state. Keywords: Temporal, retries, durable workflows, VM restarts.
Microsoft consolidates coding assistants - Microsoft is reportedly ending many Claude Code licenses and steering teams to GitHub Copilot CLI, signaling tighter ecosystem control and budget discipline even if engineers preferred the third-party tool. Keywords: Microsoft, Claude Code, Copilot CLI, consolidation.
The rising cost of AI coding - Enterprise appetite for agentic coding is colliding with token and compute bills—examples like Microsoft and Uber highlight how usage can outpace budgets, even when productivity rises. Keywords: AI cost, tokens, enterprise budgets, agents.
Qwen’s long-horizon agent claims - Alibaba’s Qwen3.7-Max is pitched for agentic workloads, including a headline 35-hour autonomous optimization run with heavy tool use—an example of models being judged on endurance, not just benchmarks. Keywords: Qwen, long-horizon, tool calling, coding agents.
Who really controls global compute - Epoch AI argues frontier labs still use only a minority of the world’s operational AI compute, with huge capacity going to inference, open models, and non-LLM workloads—though top labs may be growing fastest. Keywords: compute share, GPUs, inference, frontier labs.
AI profits, revenue, and valuations - Reports put OpenAI and Anthropic in a tight revenue race, while a separate tracker argues the broader AI sector is still collectively unprofitable—raising questions about who captures value and when. Keywords: revenue, valuation, profitability, Nvidia.
New norms for AI communication - A manifesto-style page, “Don’t quote the AI at me,” calls out the habit of pasting raw chatbot output, pushing for accountability, editing, and real human judgment in professional communication. Keywords: AI etiquette, accountability, trust, editing.
Open metadata for model comparison - Models.dev is an open, community-maintained catalog of model capabilities and metadata with an API, aiming to reduce confusion as providers multiply and model lineups change weekly. Keywords: model database, API, metadata, open-source.
Interpretability moves beyond single features - Goodfire research suggests sparse autoencoders often capture model concepts in a ‘dilution’ regime, and proposes clustering features to recover man
