Eric Roby (@codingwithroby) header
Eric Roby (@codingwithroby) profile picture

200,000+ Udemy Students | ~40k YouTube | Software Engineer | AI Nerd | Good Person to Know

Joined August 2023
Photo by Eric Roby (@codingwithroby)Photo by Eric Roby (@codingwithroby)Photo by Eric Roby (@codingwithroby)Photo by Eric Roby (@codingwithroby)Photo by Eric Roby (@codingwithroby)Photo by Eric Roby (@codingwithroby)Photo by Eric Roby (@codingwithroby)Photo by Eric Roby (@codingwithroby)Photo by Eric Roby (@codingwithroby)Photo by Eric Roby (@codingwithroby)
I’m still confused by the whole “vibe coding” concept. When I’m working with Claude or Codex, I push back constantly, especially during the planning phase. We go back and forth on architecture, product decisions, and implementation before any code gets written. Ironically, I spend far less time reviewing the code itself than I do challenging the plan. So I’m curious… Do vibe coders just let the AI rip and hope the database schema, migrations, architecture, and everything else work out? Or is “vibe coding” mostly being used to build demos and static websites rather than real production applications?
862
47
38
1,720
347,201
Six months ago, I thought the end game for AI coding was making everyone a 10x engineer. Then I watched DHH's Rails World keynote, and his 1,000x argument really made me rethink that. He was comparing the weakest programmer working without AI tools with the strongest programmer working with AI tools. He wasn't presenting a measured 1,000x result. But as an argument about how large the gap between those two extremes could become, and I kind of agree with him. It's made me question whether 10x is even the right scale for how I'm thinking about this anymore. I don't have a number I can prove. I do think I was underestimating how much this could change what one person can take on.
1
18
A lot of AI-built applications are starting to look familiar to me before I even use them. I'll look at a screen and think, this feels like the same design direction I've seen from Claude or Codex before. That's one reason I use Refero MCP when I'm working on designs. I want to give the agent actual references and work through a visual direction instead of leaving all of those choices to it. Then I can try variations, work on the design in something like Claude Design, and settle on what I want before moving into the implementation.
1
24
I'm leaning more toward modular monoliths when building with agents. I don't feel like I need to recreate a human team's frontend and backend split just because I'm running multiple agents. I'd rather give an agent a piece of work that can involve both, then use other agents for QA, testing, and review. In Claude Code, subagents work in their own context windows. I like that as a way to give each one a focused job without trying to keep every part of the process in one conversation. Then I can bring in a different models to check the result too.
13
25
431
I just passed the Claude Certified Architect, Foundations exam. I went into it thinking I already understood a lot of this stuff because of how much time I spend with AI engineering and coding agents. But I learned quite a bit. There were ideas I thought I had figured out that I ended up understanding differently while working through the exam rubric. A lot of it is pretty technical, so I'm not going to try to fit the whole thing into this post. But I'm glad I did it. If you're thinking about taking it or have questions about my experience, feel free to reach out.
65
YouTube is hard man ... but so thankful I just hit 40k subs :)
1
2
80
Nvidia needs a CEO with an engineering brain or they will be toast
1
61
The breathing sounds in AI voice conversations are kind of trippy to me. When I use ChatGPT voice, I've heard what sounds like someone taking a breath. I know I'm talking to software, but those little human details make the interaction feel strange. I'm not sure I want it to feel that human. I think I'd rather have the Iron Man version of this. An AI sidekick that helps me do things and clearly feels like an AI sidekick. I want it to be useful and easy to talk to. I don't really need the experience to feel like I'm having a phone call with another person.
1
1
21
One of the hardest parts of software right now, for me, is questioning whether I'd still design an application the way I did before agents. I'm used to building the screens and the specific workflows for creating, updating, and moving data around. But I keep thinking about applications where an agent handles more of that work behind the scenes, and the UI focuses on showing me the result and letting me interact with it. I guess the CRUD operations are still there. What changes is how much of the workflow I have to design as a fixed sequence of screens and actions. Agents are changing the entire business logic section of current apps
2
2
82
I'm still trying to understand the economics of a local AI setup. If I'm putting $10,000 - $20,000 into hardware, I want it to easily make sense But people who talk about local AI seem to only compare their hardware to frontier models. What we don't talk about a lot is how powerful the cheap frontier models are, like Luna and Haiku, compared to local AI
2
1
40
I think developer interviews are going to look very different in a few years. I'd be more interested in giving someone a few hours with an ambitious goal, letting them use AI, and seeing how they work through it with another developer. How do they explain what they want? What do they do when the agent goes in the wrong direction? Can they tell whether what came back actually meets the goal? You'd get to see their communication and their technical judgment while they're building something. I think that kind of interview could replace a lot of the LeetCode portion of hiring. System design would still have plenty of room in the conversation, but I'd want to see how someone uses these tools on a real problem.
8
10
430
I think taste is going to matter more and more as AI gets better at building software. The design, the workflow, and how the application feels when someone is using it. Those are the things I think will separate products that otherwise have very similar capabilities. Software was already about those things. I just think they'll become an even bigger part of why someone chooses one product over another. If we're all getting help with implementation, I want to spend more of my attention deciding what a good experience should look like. That still feels like a huge part of the work. An agent can give me a design, but I have to decide whether it's a design I want someone to use.
1
2
97
I think AI is going to put pressure on software teams in two directions at once: fewer people and shorter deadlines. But I also think the products we're expected to build are going to get much more ambitious. What we consider a large project now might feel very different in a few years. Omarchy is one of the projects that makes me think about that ... I expect more people to look at something that previously felt too large to take on and decide to try it with agents. That's the part I'm excited about. I'm less sure how well companies will handle these new expectations
1
2
23
Eric Roby retweeted
1/ What if an LLM could not just edit a model's weights, but re-architect it as it trains? We tried just that. The resulting architectures exit the known territory of transformers: the coding agents spliced code directly into the network. On nanochat, the loss fell below ordinary training after each new code layer was introduced. On algorithmic reasoning, accuracy rose from 55% to 84%. We even let DeepSeek-V4.1-Flash act as the agent modifying itself, raising its AIME and HMMT 2026 score from 64% to 71%. To do all this, the agents built a set of tools we’ve never seen before, which we call mechanistic intervention! Blog: doubleai.s.gy/llm-redesign
60
110
41
967
87,354
I think we're going to see more developers owning an entire application, or a much larger part of one, with agents helping them do the work. A handful of people might be able to take on something we'd previously have split across a much bigger team. That's a prediction, but it changes how I think about working with other developers. I expect more responsibility to sit with each person, even if fewer people are involved in the implementation. I still think we'll need plenty of human communication. I can also see agents handling more of the routine updates through connected tools, including drafting Slack messages about what's changed. People would still need to agree on where the product is going and how their work fits together.
1
26
I think one of the strangest parts of AI coding is being able to build something faster than you've fully decided what it should be. I'm used to implementation taking enough time that I could keep thinking through the product while the work was happening. Now I can give an agent a substantial task and let it work without being hands-on the whole time. That changes where I need to spend my attention. I need a much better understanding of the product. What we're trying to accomplish, how the feature should work, and whether the next thing I'm asking for makes sense. I think this is why the job is starting to feel more like product engineering to me. The implementation can move quickly, but I'm still responsible for figuring out what we're trying to build.
1
1
27
I used to default to web apps because I wanted my tools available when I was away from my computer and it was what I was familiar with But building native apps has changed how I think about that. For my own tools, I like having more control over how they work with the machine and where I can improve performance. A Mac app doesn't have to go through the Mac App Store. And on iOS, TestFlight gives you a way to test your own app on your phone, with each build available for up to 90 days before it expires. I have my internal setup connected through Tailscale. I kind of assumed wanting mobile access meant everything should be a web app.
1
166
I started building all my internal tools as separate web apps on different ports. Then I was talking to Adrian who said he now builds native apps, and I realized I hadn't even considered doing that. I'd never built native apps before. It just wasn't where my brain went when I wanted to make something for myself. With AI, that changed. I took design ideas from Claude and ChatGPT and built my own internal app around what I needed, including connecting it with applications already on my machine. It's funny how much of my original decision was just familiarity. I knew web apps, so every internal tool was going to be a web app.
2
2
119
I always work in a separate worktree, so if something is off I can just scrape it If I realize we're going down the wrong path, I'll stop the agent, get rid of that attempt, and start a new one with more information about what I actually want. I'd rather give a fresh agent a clearer task than keep trying to steer an implementation I already don't want.
1
38
Not why the top headlines keep saying “Karpathy built YouTube in 2 Hours” He built a UI to add his own browse list and removed thumbnails (which idk if that’s even a good idea) Everything else still uses … YouTube
1
60
I really need to automate this ... Sometimes I’m working through a plan with an agent and it keeps asking me questions with a recommended answer. And I keep choosing the recommendation. After a while, I start wishing it would just use that recommendation by default Most of the time, that’s what I want. I’m asking the agent to help me figure this out, so I’m usually happy to proceed with the approach it thinks makes sense ...
31