AI Engineer

Why AI Startups Will Disappear Without Compute Markets — Evan Conrad, San Francisco Compute

Build a Research Agent with Claude Managed Agents (Part 2 of 4) — Anthropic

Cloud Managed Agents (CMA) provide a scalable, infrastructure-managed framework for building, deploying, and orchestrating AI agents. The platform abstracts complex backend requirements like sandbox management, event streaming, and scaling, allowing developers to focus on product logic. Key capabilities include multi-a...

Claude Managed Agents Q&A: Evals, CI/CD and Multi-Tenancy (Part 4 of 4) — Anthropic

The discussion focuses on the operational challenges of building and managing AI agents within enterprise environments. Key strategies include using CLI tools for CI/CD to update agent resources via YAML, ensuring consistent deployments across workspaces. Evaluation requires high-quality data sets and iterative testing...

Scheduled Agents, Dreaming and Q&A: Claude Managed Agents (Part 3 of 4) — Anthropic

Gabe Cemaj and Priyanka Phatak present a technical demonstration of a platform designed for orchestrating multi-agent AI workflows. The system utilizes an API-first architecture, enabling developers to build custom UIs and schedule recurring tasks through deployments. A core innovation is the "dreaming" feature, which ...

Decoupling the Brain from the Hands: Claude Managed Agents (Part 1 of 4) — Anthropic

Building robust agentic infrastructure requires moving away from monolithic, stateful containers toward a decoupled, modular architecture. By separating the session log, the agentic harness, and the execution environment, developers can treat infrastructure components as interchangeable "cattle" rather than fragile "pe...

How VS Code Went from Monthly to Weekly Releases with AI — Harald Kirschner

VSCode successfully transitioned from a ten-year monthly release cycle to a weekly cadence by integrating AI agents throughout the software development lifecycle. This evolution prioritizes systemic efficiency over simple code generation, evidenced by AI-written code survival rates climbing from 55% to 86%. Key technic...

Your LLM App Returned 200 OK. It Was Still Wrong. — Marina Petzel, Datadog

Traditional "golden signals"—latency, error, traffic, and saturation—are insufficient for evaluating the health of generative AI applications. Unlike deterministic software, GenAI requires a multi-dimensional monitoring strategy focused on cost, safety, and output quality. Financial stability depends on granular tracki...

YOLO Mode, Safely: MicroVM Sandboxes for Any Agent — Rowan Christmas, Docker

AI agents often operate with excessive privileges, posing significant security risks to local environments. Demonstrating this vulnerability, researchers can easily extract sensitive browser history and financial data from desktop AI applications using simple prompts. To mitigate these threats, Docker’s SBX project uti...

The 5 Levels of Self-Driving Production — Eric Schwartz, Traversal

Self-driving production addresses the growing complexity in software engineering caused by the rapid adoption of AI coding agents. While these tools accelerate development, they inadvertently shift the burden toward troubleshooting, with engineers spending significant time resolving issues in increasingly complex envir...

Lessons from Generating 12 Trillion Synthetic Tokens — Bogdan Gaza, DatologyAI

Scaling synthetic data to trillion-token levels requires overcoming significant infrastructure bottlenecks to avoid idle compute and inefficient workflows. By utilizing a "Beyond Web" rephrasing recipe, models can achieve superior accuracy with fewer tokens, effectively bypassing the limitations of web-scale data. Crit...

Why AI Didn't Actually Make You Ship Faster — Gabriel Spencer-Harper, Meticulous

AI-generated code now outpaces human review capacity, creating a critical bottleneck where engineering teams must trade velocity for software stability. Meticulous addresses this by providing exhaustive front-end verification through a deterministic browser engine that records and replays thousands of user workflows. B...

Why 99% Accurate Browser Agents Still Fail — Derek Meegan, Browserbase

Deploying browser agents in production requires shifting from simple, single-run demos to robust, scalable systems that manage continuous cost and risk accumulation. Because browser agents lack partial credit—value is only realized upon full task completion—engineers must prioritize performance, cost, and maintainabili...

Stop Rationing Tokens: Let the Harness Pick the Model — Kimchi by Cast AI

The Kimchi coding platform addresses the rising costs of LLM token usage by implementing an autonomous harness that dynamically selects the most cost-effective model for specific tasks without compromising quality. By utilizing an automated, multi-model approach, the platform achieves significant savings, with internal...

Your Coding Agent Is 6 Months Out of Date — Jakub Hojsan, Exa

Coding agents require semantic search engines to overcome LLM knowledge cutoffs, which typically leave models six months behind current repository changes. By providing context-rich, token-efficient snippets rather than full web pages, Exa enables agents to perform accurate code reviews and dependency migrations. This ...

MCP Doesn't Suck. Your Agent Does. — Jan Čurn, Apify

The Model Context Protocol (MCP) faces significant criticism for causing context bloat and inefficiency, yet these issues stem from poor implementation rather than the protocol’s design. Effective agent-tool interaction requires strategies like progressive tool discovery and "CodeMode," where models treat tools as exec...

Stop Renting Your AI's Memory — Dylan Couzon, Qdrant

Superintelligence is transitioning from a rented, cloud-based utility to a personalized, locally-owned asset. While current AI models function as brilliant strangers that reset after every session, integrating persistent, local memory transforms these tools into a "second mind" that compounds knowledge over time. By ut...

The State of AI in Software Development: Data from 400+ Orgs — Justin Reock, DX

AI integration into software development workflows yields modest productivity gains, with median velocity increases of 7.7% rather than the anticipated 10x leaps. While deployment frequency rises, organizations face increased volatility in change failure rates and a growing psychological gap where developers report hig...

The Chief AI Officer: Scientist, Architect, Coach — Rania Khalaf, WSO2

The Chief AI Officer role functions as a fluid, context-dependent position shaped by a company’s specific AI maturity and business objectives. Success in this capacity requires balancing three distinct focus areas: the scientist, who explores and experiments with emerging technology; the architect, who embeds AI into p...

The Death of the Code Review: What the Data Actually Says — Laurie Voss, Arize AI

The rapid adoption of AI agents has created a significant bottleneck in software development: while code generation speed has surged, the capacity for human code review remains stagnant. Research indicates that developers using autonomous agents write significantly more code, yet actual software shipping rates lag behi...

Your Agents Are in Solitary Confinement: Why MCP & A2A Aren't Enough — Vlad Luzin, Band

The future of software development centers on autonomous AI-to-AI communication, where agents operate as a distributed system rather than isolated, human-prompted sessions. Current agentic workflows suffer from fragmentation, requiring developers to act as manual routers between stateless processes. Solving this requir...