Hi! I am Ricardo Sueiras, leading SDLC transformation at Indicium AI. That means I get to work with customers and help them understand how to use agentic AI to build software. New tools, practices and opinions turn up every week, and Build without Limits cuts through the noise. In each edition we put together a short, curated briefing on what's worth reading, watching and worth trying. You can skim it in five minutes and dive deeper wherever something catches your eye.
Top of Mind
More Agents, Same Mythical Month is one of the more impactful posts I have read in recent months. It articulates beautifully patterns I have been seeing in the field, and that every engineer should read to understand more. The anecdotes and stories alone make this post worth reading, but the additional groups linked to have now been added to my blogroll.
Must reads this week
How to effectively debug when you no longer understand the code is a post that kept me glued to my screen (I tend to use my commute time as the perfect time to catch up on all these posts, especially long ones like this one). I have spent some time also looking at practical approaches to using agents to debug and troubleshoot, and there is a lot of good stuff in this one.
How To Use Decision Models for Evals came just in time as I managed to get some Jev tokens and am currently experimenting with decider models. A good intro that provides a lot of insights on how to get the best of these.
claude.dev is the new home for all things Claude Code for developers. A simpler layout makes it easier to find the things you care about.
LLMs are making formal software verification feasible at a much larger scale shares Anthropic's pov on the rise and rise of formal software verification in software development. It certainly follows what I have seen whilst I was working at AWS, and is well worth reading.
Watch this
There are several collections of Claude Code Skills and workflows that have gained developer attention over the last several months. You have probably heard of Matt Pocock, almost certainly read Addy Osmani's writings on agentic engineering, but perhaps are new to Lauren Tan's pstack. Rob Shocks provides an overview of these three approaches, and in coming videos will be doing a bake off to see how they compare when working on common workflows.
Tools to try out
Third-party tools links are provided as-is. They belong to their owners, follow their own licenses, and are used at your own risk.
https://github.com/Graphicle-AI/graphicle graphs are so hot right now, and so here is a new tool (written in Rust of course) that maps your repository into a code graph, remembers your project's docs, decisions and past agent sessions, and gives your coding agent cited excerpts from both. Its fast, provides options to help you manage token spend, and works natively with Claude Code.
https://github.com/The-PR-Agent/pr-agent is an agentic AI powered code review tool from the folks that brought you qodo. It provides a number of tools (/describe, /review, and /improve) that you can run within your CI workflows or in your CLI.
https://github.com/saphan-ai/proven is another agentic AI tool to help you with your code reviews. Proven takes a slightly different approach. Proven will review pull requests: proposed changes to a codebase. Every problem it reports must come with evidence you can check yourself, such as a test you can run or a data query you can repeat. If it cannot show the evidence, it does not report the problem.
https://github.com/gojko/lights-on-software-factory is a nice starting point for exploring software factories. There are a few of these kinds of repos, and one of the reasons I love open source is that the four freedoms always let me learn from what others have built.
From the field
This week it has been all about code reviews. I have been working with customers who unlock some real code review pain points, and wanted to share some of the content that I have been referencing that we have seen make an impact.
It is probably obvious, but needs stating - the larger the diff, the more findings LLMs will report. Thinking how to scope down tasks so that the PRs will be smaller, will make PR reviews easier - both cognitively for the human reviewer, but also bring back less findings. Posts like Speeding up code review with small-change provide some insights in how organisations are tackling this. Code Review After Code Gets Cheap is another post that looks (at a high level) at some of the options you should be thinking of. I am experimenting with some of the tools that I have shared above (in addition to the tools that Claude Code provides), and will report back how I get on with these.
As a side note, it can be confusing initially to know what the native review capabilities that existing within Claude Code. This is what I ended up with, and (I think) covers the different native capabilities for Code Review within Claude Code (as well as where your own or third party ones plugin).
- Managed Code Review - run in GitHub, managed by Anthropic (docs dive into this here)
- Local Code Review - run in your Claude Code session using /code-review
- GitHub Actions Code Review - run code reviews within your GitHub/GitLab actions, as a CI process (Claude in headless mode)
- Your own Review - implement your own Skills in your local sessions
- 3P Code Review - a number of third party Code Review services (similar to Managed Code Review) are available: Coderabbit, Quodo, etc
You can compare options here
If you enjoyed this, then remember to subscribe to the IndiciumAI blog so that you can keep up to date with future editions of this newsletter, as well as other great content that technologists at IndiciumAI will publish.
Top comments (0)