Tiffany Rea

Progress Update: Accelerating Onboarding with Coding Agents Over the past two months, our team of 20+ engineers has been exploring how a coding agent can improve the way we work. The results so far have been promising—new hires are onboarding into complex codebases faster, knowledge is spreading more quickly across the team, and repetitive tasks are being handled by the agent, freeing engineers to focus on higher-value work. We’ve learned that quality improves significantly when the agent is given strong example code or best practices to work from, especially for complex enterprise level design (such as mono repo) that aren’t typically represented in model training data. Just as importantly, we’ve found that the most effective way to work with an agent is to treat it like a teammate: ask curiosity-driven questions, explore “what if” scenarios, and brainstorm together toward alternative solutions. As our coding speed has improved, a new question has emerged: how do we ensure that system design quality and technical documentation keep pace? To tackle this, I’ve proposed exploring a “document agent” that can help capture design ideas, record decision points, and integrate relevant industry research into our dev process. By pairing a coding agent for execution with a document agent for design and knowledge sharing, we aim to not only work faster but also build smarter, more resilient systems. Stay tuned as we continue to push the boundaries of intelligent collaboration.

71

2 Comments

I'm excited to share that our client-platform team at Docusign is open-sourcing our micro-frontend framework and development tooling ecosystem under the moniker 1FE <https://1FE.com>. 1FE is designed for large scaled teams who need to deliver components and widgets independently with velocity, but with a common customer experience and a shared set of tools including a CLI, and a best-practice out-of-the-gate CICD and QA pipeline. Because experiences are stitched together at run-time, individual developers can run and test their in-progress work within the context of a live production instance without impacting the broader group of customers. It was originally designed with the intention that it should be "easy to check-in good code and hard to check-in bad code". If you are an front-end engineering team struggling to release at scale and with velocity, I encourage you to check it out. #opensource #react #QA #javascript #frontend #microfrontend

248

10 Comments

Today's link dump focuses on vibe coding, but also .. on how these agents actually work under the hood. First off is this post [0] by some folks at Hiro on their approach to vibe coding. It's a great starter for folks new to this space. I liked one of the prompting questions in there: "Through discussions with other engineers and leaders at AI startups, we know that we are not alone here. Everyone faces this question of “how much AI coding is okay”. Tab completions are obviously fine, but what about letting agents run amok? What about vibe coding entire features?" Switching from there to one individual's thoughts on the topic [1], I saw an interesting contrast between how people approach this in corporate vs individual settings (and tips and tricks on how to use these more effectively). Knowing exactly what you want is important, otherwise you'll hit this: "True story- at the end of a long session, I asked Claude how my design was. It was effusive in its praise, pointed out all the things we had done for modularity, separation of concerns, etc. Then I exited Claude and restarted, and asked the exact same question, only this time it had no context. It told me that the design needed a lot of work, complained about the stuff it had just praised, etc." At the same time, as pointed out in [2], AI is an unreasonably effective partner in crime when you want to prototype/bounce your ideas off of someone and it's too soon to hit up a human for tips. I mostly found myself nodding along with this piece, though ... I'd disagree with the "surprisingly" at the start of this quote: "Surprisingly, I think that the fact that I need to plan my features carefully before rushing into implementation is making me a better developer overall. This happens simply because it forces me to document the implementation and think it through before jumping into code. I also find myself explaining my reasoning more clearly because I have to write it down for the AI" Beyond using these agents, though, I've found it to be a great learning experience understanding how all of these tools work under the hood. [3] is a great foray into understanding why/how Claude Code works the way it does: "The main takeaway, again, is to keep things simple. Extreme scaffolding frameworks will hurt more than help you." ... and if you'd like to go even further, you can build your own agent in under 300 lines of code if you follow the advice in [4]. I highly recommend that last piece, because it's chock full of advice in addition to just a great tutorial. "In my opinion, any disruption or job loss related to AI is not a result of AI itself, but rather a consequence of a lack of personal development and self-investment. If your coworkers are hopping between multiple agents, chewing on ideas, and running in the background during meetings, and you're not in on that action, then naturally you're just going to fall behind."

36

8 Comments

Hand-tuning a system to achieve great performance is an anti-pattern. There should be no need to do so. Software exists to solve a problem for you; why should you have to tell it what's the best way for doing so? Instead, systems should be designed to automatically find the right configuration for things like number of elements in a batch, queue and buffer sizes, cache retention, etc. While it's an interesting question whether AI can help with that task, in most cases this isn't needed; classic control theory can do the job, fast, precisely, and cost-effective. As a maintainer of a library, and even more so of managed services, you should push back hard on every new knob, every additional setting, every new parameter for letting the user configure implementation details like that. It substantially reduces usability. It means pushing work to the user instead of having the software do it.

77

29 Comments

eStreams's emoji processing service is overwhelmed with 100k/sec incoming requests? 😵 Triggering cascading failures across your entire system, bringing everything down. 💥 Solution: Back pressure A feedback mechanism that allows a slow consumer to tell a fast producer to "slow down." Here’s how it creates stable, resilient systems 1. Downstream service processing the data, signals that its internal buffer or queue is nearing capacity     2. Upstream service receives this signal and temporarily stops or reduces the rate at which it sends new data     3. This gives the consumer crucial time to catch up on its work without being overwhelmed and crashing     4. Once the consumer has processed its backlog and freed up capacity, it signals the producer that it's ready for more data     5. It effectively changes the dynamics From the producer blindly pushing data; to the consumer pulling data only when it has the capacity to handle it 🤔 over to you, what strategies have you used to handle uneven data flow between your services? Let me know in the comments! 👇 --- You've got System Design skills but showcasing them in 60 minutes interview is a completely different challenge for e.g. where to start? How much time to spend on HLD vs LLD etc Let me share a FREE 7-day email course that provides a step-by-step framework for acing your 60-minute System Design interview 📣 Join 1500+ subscribers and sign up for the course at systemdesign.engineer, to learn time allocation to demonstrate full range of design skills

42

15 Comments

Safety tips for devs & vibe coders diving into Gemini CLI: - The CLI will ask for permissions. Read it. Review it. - gemini --sandbox is your safe playground (macOS Seatbelt, Docker and Podman support) - gemini --checkpointing = instant "undo" button with /restore - Back up, lock down, and breathe before smashing allow always Enjoy ❤️

84

3 Comments