Skip to content

Tech News

Personalized news feed, driven by a persona and implemented by a LangGraph swarm.

Available for work Applied AI Player / Coach available for startup-like teams. Build with me →
AWS Open-Sources Emulator for Local Testing of 200+ Services by Coding Agents
AI/ML / infoq.com

AWS Open-Sources Emulator for Local Testing of 200+ Services by Coding Agents

AWS open-sourced AWS Emulator, a local HTTPS endpoint written in Rust that emulates the public APIs of over 200 AWS services, generated from 400+ Smithy API models and built entirely by coding agents. It supports the AWS CLI, SDKs, and an embedded mode for Rust tests, but does not evaluate IAM policies, uses a single shared account per process, and has in-memory-only state. Fidelity varies by service: S3, DynamoDB, and SQS pass about two-thirds of published tests, while CloudFormation passes under one-fifth, and AWS publishes versioned per-service fidelity scores with each release.

Why it matters

For platform engineers and AI agent developers, this provides a first-party, agent-built emulator with official API fidelity scores, enabling local iteration without cloud costs, but its security and state limitations mean it cannot replace LocalStack for permission testing or infrastructure-as-code pipelines.

Windows 11 is collecting data on which coding agents and developer tools you use
AI/ML / xusheng.dev

Windows 11 is collecting data on which coding agents and developer tools you use

Reverse engineering of Windows 11's MissionControlAggregator.dll reveals it tracks usage of coding agents (codex.exe, claude.exe, copilot.exe, cursor.exe) and developer tools (python.exe, dotnet.exe, gcc.exe, code.exe) via ETW events, then reports this data to Microsoft's optional diagnostic telemetry (enabled by default). The DLL subscribes to KernelProcess.AppStateChangeSummary and Win32kTraceLogging.AppInteractivitySummary for Windows apps, and Microsoft.Windows.Lxss.Manager.Exec for WSL, using executable-name matchers with publisher verification for Node.js, Python, and .NET tools. Feature ID 62580080 (CLICodingAgentUsage) gates the new agent checks, and data uploads unless users explicitly opt out of optional diagnostics.

Uber’s Legal AI Agent Cuts Contract Review Time by 20% with Rules and RAG
AI/ML / infoq.com

Uber’s Legal AI Agent Cuts Contract Review Time by 20% with Rules and RAG

Uber's Legal Redlining Agent, built with LangGraph, React/TypeScript, and a Python backend, reduced contract review time by over 20% with 91% accuracy. The system combines RAG from legal playbooks, a deterministic rules engine for non-negotiable clauses, and a feedback loop capturing lawyer decisions to improve semantic retrieval. It uses OpenSearch for vectorized rules, Redis for session state, and exponential decay weighting (365-day half-life) to prioritize recent historical examples.

GitHub Rebuilds the Copilot App's Diff Surface to Open Million-Line Pull Requests Without Jank
Open Source / infoq.com

GitHub Rebuilds the Copilot App's Diff Surface to Open Million-Line Pull Requests Without Jank

GitHub rebuilt the Copilot app's diff surface to handle a pull request with 2,200 files and over a million changed lines without jank, splitting the document into deterministic code height and dynamic block domains keyed by identity rather than pixel position. The team abandoned a ResizeObserver-per-block feedback loop for a frame-synchronous exception model, and added structured probes and an autopilot-driven regression suite to catch performance regressions. The work responds to AI coding agents producing massive PRs that shift the bottleneck from writing code to reviewing it.

Apple reveals new Mac security controls as AI agents raise privacy risks
AI/ML / techcrunch.com

Apple reveals new Mac security controls as AI agents raise privacy risks

Apple is overhauling Mac security controls in macOS 27.3 to address privacy risks from AI agents that require full disk access. The update splits Device Control and Data Access into two distinct settings—System & App Protection Override and Accessibility—and introduces granular permission prompts, a transparency indicator for app-to-app control, and new guidance nudging users toward least-privilege access. These changes follow incidents where Meta's Muse app read private messages and a ChatGPT Mac app vulnerability exposed sensitive data, balancing Apple's growing Mac mini/Studio sales to AI developers with user privacy.

I Built a Serverless OLAP Engine in the Browser with DuckDB-WASM (No AWS, $0)
Cloud / dev.to

I Built a Serverless OLAP Engine in the Browser with DuckDB-WASM (No AWS, $0)

DuckDB-WASM enables a fully serverless OLAP engine in the browser, demonstrated by an Earthquake Explorer that processes 58,000 events with star schema, CUBE, and ROLLUP queries. The pipeline uses GitHub Actions to extract from USGS API, transform with DuckDB SQL into Parquet files, and serve statically via GitHub Pages—eliminating any cloud infrastructure costs. This approach proves that for megabyte-scale datasets, client-side analytical processing can replace traditional OLAP servers like ClickHouse or Snowflake.

Dax Raad rebuilt his browser tool in Rust for $60. Then memory use fell 97%.
Languages / thenewstack.io

Dax Raad rebuilt his browser tool in Rust for $60. Then memory use fell 97%.

Dax Raad rebuilt OpenCode Browse's browser automation tool in Rust for $60 using Claude Opus and DeepSeek, slashing the binary from 53-64 MB to 4.3 MB and cutting per-connector memory from ~190 MB to 5.6 MB—a 97% reduction. The rewrite replaces Playwright and Patchright with a WebSocket-based relay using Chrome DevTools Protocol and embedded QuickJS for JavaScript execution, though it drops time zone emulation and inherits a busy-loop eval issue. Raad now owns maintenance previously handled by those libraries, trading runtime efficiency for potential long-term technical debt.

PostgreSQL 19's pg_plan_advice turned a 28ms query into 264ms under load
DevTools / dev.to

PostgreSQL 19's pg_plan_advice turned a 28ms query into 264ms under load

PostgreSQL 19's new pg_plan_advice contrib module lets developers inject planner hints via SET commands, but testing shows it can degrade performance dramatically when misapplied. A 28ms indexed query ballooned to 264ms under 10-client concurrency after forcing a sequential scan, revealing a 9.3x penalty versus 2.6x single-client. The module provides honest feedback when hints can't be honored, but the author found no case where hints improved on the planner's default choices.

Strands to Strands over A2A on Bedrock AgentCore, With IAM as the Only Credential
AI/ML / dev.to

Strands to Strands over A2A on Bedrock AgentCore, With IAM as the Only Credential

AWS Bedrock AgentCore Runtime now supports the Agent2Agent (A2A) protocol for Strands agents, using IAM and SigV4 as the sole authentication mechanism—no API keys or OAuth tokens required. A step-by-step guide demonstrates a worker agent hosted on AgentCore with IAM inbound auth, a caller signing requests via SigV4, and a single IAM policy controlling access. The example worker exposes a currency conversion tool using European Central Bank rates, served via the `serve_a2a` function on port 9000.

AI/ML / techmeme.com

Anthropic tells investors Claude Mythos Preview got its AI running on AMD GPUs in ~3 days with "basically no human intervention", below the usual 6 to 12 months (Shane Downing/Tom's Hardware)

Anthropic reportedly achieved a major milestone by porting its Claude Mythos Preview AI to run on AMD GPUs in approximately three days with minimal human intervention, a process that typically takes six to twelve months. This suggests that AI itself may be capable of writing the low-level code needed to adapt models to non-Nvidia hardware, potentially reducing the competitive advantage of Nvidia's CUDA platform.