Blog
Notes and analysis on AI development.
Four-Nines Agents on Kafka, Confluent Bets on Real-Time AI
Confluent Intelligence Q2 turns Kafka and Flink streams into real-time context, control, and security infrastructure for AI agents.
When Flash Beats Pro, Agent Economics Take Over
Gemini 3.5 Flash is not just another fast model release. It points to the cost, latency, and routing fight behind coding agents and AI search.
When the search box becomes a 24-hour agent, Google tests mini apps
Google Search agents extend search from answers into persistent monitoring, task state, alerts, and generative UI mini apps.
Gemini API Managed Agents turn model calls into sandboxed workers
Google Gemini API Managed Agents move model calls into isolated Linux sandboxes with stateful agent execution.
Free Gemini CLI gets a deadline as terminal agents move to Antigravity
Google is moving the personal and free Gemini CLI path to Antigravity CLI. The June 18 cutoff marks a shift in the operating layer for coding agents.
The switch to review before August 17, Atlassian AI data contribution
Atlassian data contribution settings show how Jira, Confluence, Rovo, and Teamwork Graph data defaults now shape AI improvement loops.
Codex Goals and the New Completion Contract for Coding Agents
OpenAI Codex Goals turns long-running coding work into an evidence-based loop with objectives, verification surfaces, constraints, and budgets.
KPMG puts Claude inside the tax and legal workbench
Anthropic and KPMG are turning Claude into an agent layer for Digital Gateway, private equity modernization, and Big Four delivery.
11 Seconds of Audio in Under 8 Seconds, Without a GPU
Google and Arm show how on-device generative AI is moving from model releases into CPU runtimes, quantization, memory limits, and silicon features.
Composer 2.5 shows Cursor training for reward hacking
Cursor Composer 2.5 shows the coding-agent race shifting from benchmark scores toward long-task failure points, targeted feedback, and reward-hacking detection.
Google AI Overviews exposes the gap behind citation cards
A May 13 arXiv study measured 55K Google searches and 98K AI Overview claims, showing where citations, ranking, and publisher economics diverge.
ECHO makes stderr part of the coding agent world model
Microsoft Research ECHO turns terminal output into a direct learning signal so coding agents can learn from failed logs, not only final rewards.