Tech Radar

Signaux technologiques pour systèmes réels.

Des mises à jour sélectionnées avec le jugement d'UniQubit : ce qui a changé, pourquoi les constructeurs devraient s'en soucier et comment cela affecte les produits réels.

platform_update15 mai 2026

Hidden Bottleneck in ClickHouse Query Planning Identified by Cloudflare

Cloudflare traced a major slowdown in its billing pipeline to hidden lock contention inside ClickHouse’s query planning layer after a partitioning change altered execution behavior at scale. Traditional database metrics did not clearly expose the issue, which was ultimately identified using flame graph analysis. The fix involved replacing an exclusive lock with a shared lock in the query planner to reduce contention under heavy concurrency.

#clickhouse#cloudflare#query-planning#lock-contention#database-optimization
Editorial abstract cover for Hidden Bottleneck in ClickHouse Query Planning Identified by Cloudflare, showing a clickhouse and cloudflare update as connected infrastructure signals
developer_tooling15 mai 2026

Mobile Management of Coding Tasks with Codex

OpenAI is bringing Codex into the ChatGPT mobile app, allowing developers to monitor, steer, and approve coding tasks remotely across devices and environments. The rollout includes support for Remote SSH and Hooks, enabling automation workflows and remote execution management from mobile. Device synchronization is handled through a relay-based architecture designed to avoid directly exposing development environments to the public internet.

#codex#mobile-app#remote-work#automation#security
Editorial abstract cover for Mobile Management of Coding Tasks with Codex, showing a codex and mobile-app update as connected infrastructure signals
security_plan14 mai 2026

Custom Sandbox Enhances Codex Security on Windows

OpenAI developed a custom secure sandbox for Codex on Windows, addressing limitations of existing tools like AppContainer and Windows Sandbox. It restricts file writes and network access using Windows SIDs and write-restricted tokens, operating without admin privileges. The result is a constrained execution environment for AI coding agents that preserves user-level permissions while reducing operational risk.

#sandbox#windows#codex#security#openai
Editorial abstract cover for Custom Sandbox Enhances Codex Security on Windows, showing a sandbox and windows update as connected infrastructure signals
pricing_change13 mai 2026

GitHub Copilot Individual Plans: Transition to Usage-Based Billing with Flex Allotments

GitHub will transition Copilot individual plans toward usage-based AI consumption starting June 1, 2026. The update introduces flex allotments for Pro and Pro+ users alongside a new Max tier designed for sustained, high-volume workflows. Unlimited code completions and edit suggestions remain included on paid plans, while heavier AI usage increasingly maps to compute-style consumption.

#github#copilot#billing#plans#ai
Editorial abstract cover for GitHub Copilot Individual Plans: Transition to Usage-Based Billing with Flex Allotments, showing a github and copilot update as connected infrastructure signals
research12 mai 2026

How Enterprises are Scaling AI: Trust, Governance, and Workflow Design

European enterprises are scaling AI by treating it as an operational layer rather than a standalone productivity tool. Organizations including Philips, BBVA, and Scania are embedding governance, security, and workflow redesign directly into AI adoption strategies, with human oversight remaining central to deployment.

#ai#scaling#trust#governance#workflow-design
Editorial abstract cover for How Enterprises are Scaling AI: Trust, Governance, and Workflow Design, showing a ai and scaling update as connected infrastructure signals
security_plan11 mai 2026

Running Codex Safely at OpenAI: Security and Compliance Focus

OpenAI has introduced a security-focused deployment model for Codex that emphasizes sandboxing, network controls, approval workflows, and operational telemetry. Codex operates inside constrained execution environments where low-risk actions can be automated, while higher-risk operations require explicit human approval. Authentication is managed through ChatGPT enterprise workspace controls, and OpenTelemetry export provides organizations with detailed audit visibility into agent behavior.

#openai#codex#security#sandboxing#telemetry#network-policies
Editorial abstract cover for Running Codex Safely at OpenAI: Security and Compliance Focus, showing a openai and codex update as connected infrastructure signals
developer_tooling8 mai 2026

Token Efficiency Optimization in GitHub Agentic Workflows

GitHub is treating token consumption in agentic workflows as an engineering performance problem, not just an API billing issue. Its Agentic Workflows run as GitHub Actions for maintenance and CI tasks, operating against real API rate limits. In April 2026, GitHub began normalizing token usage data across workflow runs to identify cost drivers and optimize repeated agent automations.

#github#workflows#token-efficiency#automation#optimization
Editorial abstract cover for Token Efficiency Optimization in GitHub Agentic Workflows, showing a github and workflows update as connected infrastructure signals
developer_tooling7 mai 2026

Evaluating Agentic Behavior in AI Models for Developer Tools

GitHub is highlighting a key challenge in AI coding agents: validating agentic behavior when “correct” is not always deterministic. As tools like GitHub Copilot move beyond code completion into more autonomous workflows, the reliability problem shifts from checking final outputs to evaluating the agent’s reasoning, decisions, and behavior across complex development tasks.

#ai#github#copilot#developer-tools#agentic-frameworks
Editorial abstract cover for Evaluating Agentic Behavior in AI Models for Developer Tools, showing a ai and github update as connected infrastructure signals
platform_update6 mai 2026

Choosing Between Supabase Realtime and ETL for Data Handling

Supabase separates Postgres change streams into two operational paths: Realtime for immediate user-facing updates and ETL for reliable movement into analytical destinations. Both use logical replication from Postgres, but they serve different latency, durability, and destination requirements.

#supabase#realtime#etl#postgres#data-movement
Editorial abstract cover for Choosing Between Supabase Realtime and ETL for Data Handling, showing a supabase and realtime update as connected infrastructure signals
infrastructure6 mai 2026

Enhancing AI Training Networks with MRC (Multipath Reliable Connection)

OpenAI, in collaboration with AMD, Broadcom, Intel, Microsoft, and NVIDIA, has introduced the MRC protocol to enhance GPU networking performance and resilience in AI training clusters. Released through the Open Compute Project, MRC employs multi-plane networks and adaptive packet spraying to address network congestion and failures. This protocol indicates a strategic shift in AI infrastructure design, focusing on redundancy and failure management to support scalable AI systems.

#ai#networking#supercomputers#openai#protocol
Editorial abstract cover for Enhancing AI Training Networks with MRC (Multipath Reliable Connection), showing a ai and networking update as connected infrastructure signals
model_release5 mai 2026

GPT-5.5 Instant: Enhanced Accuracy and Personalization

The GPT-5.5 Instant update to ChatGPT's default model significantly enhances accuracy and personalization, reducing hallucinated claims by 52.5% and inaccurate claims by 37.3% compared to its predecessor. These improvements are particularly impactful in high-stakes fields such as medicine, law, and finance, as well as in image analysis and STEM question answering. The update aims to provide stronger, more concise answers with a natural conversational tone, suggesting a shift in the role of AI models from general-purpose assistants to more reliable system components.

#gpt-5.5#chatgpt#ai-model#personalization#accuracy#stem
Editorial abstract cover for GPT-5.5 Instant: Enhanced Accuracy and Personalization, showing a gpt-5.5 and chatgpt update as connected infrastructure signals
tech-radar5 mai 2026

OpenAI's Re-architected WebRTC Stack for Low-Latency Voice AI

OpenAI has announced a re-architecture of its WebRTC stack to enhance real-time voice AI interactions by addressing latency and infrastructure constraints. This update is designed to support over 900 million weekly active users, ensuring fast connection setup and low media round-trip time. The new architecture involves a split relay and transceiver setup to maintain low latency and stable connections across OpenAI's infrastructure. Signal: Voice AI shifting from feature layer → real-time infrastructure layer

#openai#webrtc#voice-ai#real-time#platform-update
Editorial abstract cover for OpenAI's Re-architected WebRTC Stack for Low-Latency Voice AI, showing a openai and webrtc update as connected infrastructure signals

Showing 12 of 28 signals