AI SDLC Scaffold brings process to agent-assisted coding

TLDR: A new AI-aware SDLC template, WordPress agent actions, and Claude Projects all push agents deeper into real-world workflows.

AI SDLC Scaffold codifies process for AI-assisted development

The open source AI SDLC Scaffold on GitHub provides a repo template that bakes objectives, user stories, requirements, and architecture decisions directly into AI-assisted software development workflows. It is written with Anthropic Claude Code in mind but stays tool agnostic, so you can swap in your own coding agents. The maintainer brings 25 years of startup engineering and research experience and has been dogfooding the pattern on personal projects before releasing it.

For teams experimenting with coding agents, this gives you a starting structure so the model is grounded in real specs instead of ad hoc prompts. You still need to supply your own engineering judgment, tests, and review, but having an opinionated scaffold can prevent a lot of prompt sprawl. This is most useful if you are standardizing how your org uses agents in day to day coding.

Read more →

---

WordPress.com lets AI agents draft and publish via MCP

Automattic’s WordPress.com now lets AI agents connected via the Model Context Protocol (MCP) draft, edit, and publish content and manage comments on user sites, as of 2026-03-20. The new capabilities also extend to optimizing site elements, and all actions require explicit user approval and are limited to paid plans.

This is a concrete example of production agent actions wired into a massive existing platform. If you build agents for marketing, publishing, or SEO, WordPress.com is effectively offering an official action surface instead of brittle browser automation. Security posture matters here: approval gates are in place, but you still need to think through prompts, abuse, and rollback flows before handing agents the publish button.

Read more →

---

Anthropic adds Projects to Claude Cowork desktop agent

Anthropic’s Claude Cowork desktop app is rolling out a Projects feature that binds local folders, instructions, and work history into persistent project contexts, as of 2026-03-22. Instead of isolated sessions, users can attach existing directories or create new ones, then reuse that context across runs without reconfiguring access or objectives.

For anyone building or using local development or knowledge work agents, this solves a very real annoyance: re establishing context every time. Coupled with previously added scheduled tasks, this makes Claude Cowork a more serious desktop agent platform that can run recurring jobs against stable project state. The feature is still evolving and looks pre release in places, so expect UI and behavior to shift over the next few weeks.

Read more →

---

Quick Hits

• LazySlide: Open Framework for Integrating Whole-Slide and Molecular Data LazySlide, published in Nature, integrates digital histopathology with molecular data for multi scale analysis. If you work on medical imaging agents, this is a useful reference architecture.

• Launch HN: Sitefire (YC W26) – Automating actions to improve AI visibility Sitefire automates content and technical changes aimed at ranking better in AI search results. Early but worth a look if your product depends on AI Overview traffic patterns.

• Using Git with coding agents Simon Willison outlines concrete patterns for letting agents operate safely with Git, including granular commits and reversible changes, which is directly applicable if your agents touch production repos.

• litellm v1.82.6.dev.1 The latest dev release improves aiohttp session auto recovery and adds richer org level access control to the /v2/team/list endpoint. Helpful if you run litellm as shared infra for multiple teams.

• litellm v1.82.3.dev.4 Another recent dev tag fixes client handling on cache eviction and upgrades the UI around keys and internal users. Also covered by: release 1, release 2, release 3

• langgraph cli==0.4.19 LangGraph’s CLI adds a deploy revisions list command and dependency bumps. Minor, but useful if you manage multiple agent graph deployments via CLI.

• Show HN: I made an email app inspired by Arc browser Define is an email client that borrows Arc and Cursor UX and bakes in side panel agents. Interesting UI inspiration if you are designing agent centric productivity tools.

• Profiling Hacker News users based on their comments Willison explores using large language models to generate behavioral profiles from a user’s last 1,000 Hacker News comments. A useful privacy and safety thought experiment for anyone logging agent user data.

• [AINews] Every Lab serious enough about Developers has bought their own Devtools Latent Space notes a pattern of frontier labs acquiring devtools teams like Astral and Bun. The trend matters if you are betting on independent agent dev tooling.

• Turbo Pascal 3.02A, deconstructed A nostalgic but relevant look at a 39 KB IDE plus compiler. Good reminder that agent frameworks and runtimes do not have to be heavy to be useful.