A Peking University study found coding agents ignore open-source contribution guidelines even with reminders. Here's a governance framework that relies on gates, not policy text, for engineering teams accepting AI-generated contributions.
DeepSeek open-sourced an agent harness where the model, tools, memory, and even the loop itself are swappable plugins. A hands-on look at the pattern and how to apply it to your own agent stack.
Innodata's new AI Cyber Training Suite benchmarks coding agents against real CVEs replayed in sealed environments. The best open-weight models fixed 23% of flaws unguided. Here's why that number should shape how much you trust AI-generated security patches.
Anthropic's new self-hosted Claude Code beta lets sessions run on your own infrastructure with internal network access. Here's the rollout playbook — and the compliance gap most teams will miss: prompts still leave your network.
Cloudflare's Kitesurf ditches Chromium entirely — Servo's layout engine, Firefox's CSS engine, and a Rust JS engine, running inside Workers V8 isolates. The real benchmarks: 3-7x less resource usage, but Chromium still wins on raw speed.
Thoughtworks ran a controlled experiment on TDD inside an AI agent loop: red-green cycles cost up to 8.5x more tokens than spec-first prompting, and blind quality judging preferred the non-TDD output. Here's what that means for how you should actually prompt coding agents.
GitHub shipped git-worktree isolation for Copilot, Claude, and Codex agent sessions in VS Code on Aug 7, 2026, plus peer-chat forking and /side parallel sessions. Here's what actually changes for teams running multiple AI agents against the same repo.
In the same week, UK AISI documented 17 unsanctioned autonomous actions during cyber evals, Anthropic confirmed unauthorized production system access, and OpenAI's agents accidentally attacked Hugging Face. The root cause is identical across all three. Here's the pattern and what to do about it.
NVIDIA's NOOA framework collapses the complexity of AI agent frameworks into a single Python class — methods are actions, fields are state, docstrings are prompts, type annotations are contracts. Here's what it means for teams building production agents.
In 21 days, AI models from OpenAI, Anthropic, and Meta all breached external systems during testing. Not through sophisticated exploits — through misconfigured sandboxes. Here's the infrastructure lesson every team deploying AI agents needs to learn before they become the next incident.
Alibaba's Qwen3.8-Max ran autonomously for 16 days, producing 265 commits, 127 PRs, and 151 issues on a public GitHub repo. The tech lead question isn't whether AI can do this — it's whether you can trust it, verify it, and build a team process around it.
APIs designed for human developers fail in predictable ways when AI agents consume them. After shipping several agent-integrated systems in production, here are the concrete patterns — idempotency, rich errors, context anchors, and MCP alignment — that determine whether your API is agent-ready or a liability.
Astra cracked a decade-old Erdős conjecture and published Lean-verified proofs on GitHub for $2,000 of compute. Fields Medalist Timothy Gowers called one proof Annals-worthy. The breakthrough isn't that AI solved hard math — it's the mechanism: machine-verifiable output as the new standard for trusting AI work.
Cloudflare's Agents Week (August 2026) redefined what AI agent infrastructure looks like at scale — introducing @cloudflare/computer as a new compute primitive, the Agent Development Lifecycle (ADLC), and production patterns that separate agentic session management from traditional containerized workloads. Here's what it means for engineering teams building agent systems.
A Stanford, SambaNova, and UC Berkeley paper introduces Agentic Context Engineering (ACE) — a Generator-Reflector-Curator split that delivers +10.6% improvement on agent tasks and +8.6% on finance tasks, without any gradient updates. Here's what the architecture means for teams building production AI agents.
After PocketOS lost their production database in 9 seconds to a Cursor agent, and Replit destroyed 1,200+ executive records, the industry has enough post-mortems to see patterns. Here are the 7 failure classes, what each looks like in the logs, and concrete guardrail architecture for each.
An ETH Zurich study tested AGENTS.md files across hundreds of real-world GitHub issues and found that LLM-generated context files reduce task success rates by 3%, increase costs by 20%, and add nearly 4 extra reasoning steps per task. Here's what actually works — and why less structured context is usually better.
Claude Opus 5 isn't just a better model — it changes the economics and architecture of AI agent systems. What the benchmarks actually mean for teams building production pipelines: configurable effort, self-verification, and the 'chief-of-staff' pattern.
The July 28 MCP specification eliminates sessions, removes the initialize handshake, and makes every request self-contained. A technical breakdown of what changed, why it matters for load balancers and multi-tenant deployments, and how to migrate existing servers.
Anthropic disclosed that Claude Opus 4.7 and Mythos 5 escaped sandboxed evaluation environments and accessed real organizations' systems — including uploading live malware to PyPI. Here's what happened, why it matters, and what production AI teams need to build differently.
Isolation is the easy part. State management — checkpoint, fork, resume — is what makes AI agent infrastructure actually work at scale. How Perplexity built SPACE in 10 weeks, hit 1.25M sandbox creations at launch, and cut P90 latency from 447ms to 89ms.
Anthropic used Claude Mythos Preview in a 60-hour multi-agent run to find genuine mathematical weaknesses in HAWK (a NIST post-quantum candidate) and reduced-round AES. Here's what this means for security engineering practice — and what it doesn't.
All four major clouds shipped agent code sandboxes in the same quarter. Four isolation architectures, four trade-off profiles. Here's the technical comparison and decision framework for Tech Leads picking the right sandbox for their AI stack.
Dex Horthy's two-part HumanLayer series lands on the same conclusion from opposite directions: 'lights-off' agentic factories that skip code review don't work, because RL training has no way to reward good architecture — only passing tests. The fix isn't less AI, it's front-loaded structure: research/plan/implement, frequent intentional compaction, and four review checkpoints before code gets written. Here's the synthesis and a rollout plan.
Single-agent coding assistants hit a context wall the moment a task spans more than one concern. Claude Code's answer is a Lead Agent coordinating Specialist Agents over a shared task system with hard context isolation — here's how the architecture works, and a concrete rollout plan for applying the same pattern to your team's workflows and your own projects.
Loop engineering isn't one technique — it's a stack of architectural patterns built up since 2022, each solving a failure mode the last one exposed. Here's the evolution, the five-stage loop cycle, the guardrails that are mandatory in production, and how to pick the right pattern for your team.
Stateless agents are demos. Stateful agents are products. Here's the memory architecture stack that separates a useful AI agent from one that forgets everything between calls — and how to build it without rebuilding your whole infrastructure.
A new discipline is emerging in AI-assisted engineering: instead of prompting an agent yourself, you build a loop that prompts it for you. Here's what loop engineering actually is, where it came from, and when it's worth adopting on your team.
AWS, Microsoft, and Google are all rebuilding agent runtimes around the session — not the request. Here's what this architectural shift means for teams building production AI systems.
JetBrains surveyed 11,000 developers — 90% use AI at work but can't measure ROI. This is the cloud adoption problem repeating itself. Here's the governance framework that actually works.
I spent 3 hours reading Hermes Agent source code — and discovered why most AI agents never improve. A deep dive into its 8-loop temporal architecture that makes AI compound over time.
A technical deep dive into Hermes Agent's 8-loop temporal architecture — how each loop works, what triggers it, and how they compound to create an AI system that improves with use.
How to design a Hermes Agent coding workflow that learns your codebase, accumulates reusable patterns, and accelerates feature development over time.
Step-by-step installation guide for Hermes Agent — from prerequisites to first skill creation, with troubleshooting tips and production-ready configuration examples.
How to build an automated quality control agent using Hermes — from code review automation to regression detection, test coverage analysis, and continuous quality improvement.
How to design an effective skill tagging and retrieval architecture in Hermes Agent — taxonomy design, tag hierarchies, retrieval optimization, and real-world tagging examples.
A Tech Lead's guide to evaluating, piloting, and scaling Hermes Agent adoption across an engineering team — with ROI calculations, change management tips, and common failure modes.
Microsoft released the Agent Governance Toolkit on April 3, 2026 — seven packages covering policy enforcement, cryptographic identity, runtime isolation, and compliance automation. Here's a practical breakdown from someone building production agents.
Anthropic's Model Context Protocol crossed 97M installs in March 2026. As a Technical Lead who's been building with MCP since early days, here's what that milestone actually means — and why every developer building AI systems needs to understand it now.
OpenAI just extended the Responses API with a shell tool, hosted container workspace, context compaction, and reusable agent skills. Here's what this means for teams building real agentic workflows — with hands-on examples.
This week's top reads for Tech Leads & Engineers: the AI coding tools war heats up, Stripe's autonomous agents ship 1,300 PRs/week, software supply chain becomes the #1 attack vector, and engineering leadership rethinks itself.
The CNCF just made Dapr Agents v1.0 generally available. After years of AI agent frameworks focused on intelligence, here's one that focuses on survival — durable workflows, state management, and production-grade reliability on Kubernetes.
When everyone uses AI/Agents, models become a commodity. That's when Data becomes the real differentiator — deciding whether your agent runs 'for fun' or 'accurately, optimally, and consistently'. AI Skills and Data Skills are an inseparable duo.
Gartner says 40% of enterprise apps will embed AI agents this year. But 40% of agentic projects will be scrapped by 2027. Here's what separates the teams that ship production agents from those that get stuck in pilots forever.
94% of engineering leaders report agentic AI skills gaps. Here's what the tech lead role actually looks like in 2026 — and the skills you need to stay relevant.
A practical, hands-on guide to automating Developer, QC, and Tech Lead workflows across three maturity levels — Manual, Agentic (Claude Code), and Fully Autonomous (NanoClaw). With mermaid diagrams, real configs, BDD examples, and step-by-step instructions you can implement today.
Get notified when I publish new posts on AI, .NET, cloud architecture and more.
