Cert is expired on that link, but yes, I completely agree. The hype on LLMs is making me start to feel insane. Yes, I ship slower, but I hope to massively outlast competitors whose designs are incoherent and which no one understands.
It all depends on the product timeline. Many have come to realize that products that want to survive long term cannot do without architects that ensure codebases don't diverge. A UI/UX genius with no understanding of code will not be able to create something like Zapstore, for example, no matter how much they spend on agents.
Recommended read: https://blog.kierangill.xyz/oversight-and-guidance
Discussion
slowing down feels right
you can have best of both worlds. I don't vibe code myself, but I can integrate vibe code submitted by others and fix it up faster than I could write it all myself.
i'm starting to view it as a bunch of background processes managed by other people that get something 80% there.
the speedup i am seeing is taking the 80% code and doing 20% of work here and there instead of 100% of everything.
Yeah, I think there is good to be had. It's just very nuanced and difficult to manage without spinning out or producing only the appearance of value.
Everyone has such nuanced perspectives on this it's great to hear. As a software developer myself almost exclusively vibe coding for the past 2 months, here are my thoughts.
I was able to be productive with one hand. Since I had twins, I had an infant in my arms for 75% of the time since they were born, but was still able to create things while doing that.
Llms are JV. They will not replace senior engineers, but they do take the place of a jr dev, the same way they take the place of a paralegal. Organizations still need to hire those rolls to train people up, but one man shops can get 80% of the way to the solution faster.
We still live in a fiat world where time to market is important, so getting a quick and dirty project out the door is highly valuable to determine whether or not your idea has a market or if you're just building a tool for yourself. Building tools for yourself vs others is dramatically different way of coding. Once you e determined a project has value outside of your hands, you can spend time and money honing it.
The chat based ui for llms is not going to enhance products. I'm building a tool called vibe check, where you have your typical kanban workflows. Having llms work off of cards or issues will yield much better results that stupid chats. Especially if I get to the point where I can have specialized agents for certain functions (a QA agent, a security minded agent, a ux designer, a test automation agent, etc) then you can have work more fully fleshed out, then have a human in the loop review it and request changes before having the coder (agent or human) go back and make changes. This is how agile development has worked in the industry. Like imagine a user story that when it gets to the done column has been fully QAd, is deemed secure, has good design, etc and there isn't coordination required between so many people, just one human coder? Enhancing a good programmer with a security/QA/devops/designer will allow teams to augment their skills until they have time or money to develop their team or org to hire people with those skills
I would try this, sounds very interesting
yeah the vibe kanban thing is a great idea
I like this idea.
I’ve seen someone running like 4 teams of AI, each team with 3-4 agents to achieve specific goals. Almost like how a company would be structured.
I don’t know if that is really needed in that way, but I find it very interesting. Where you will have a QA team, planning team, front end team? Infra teams. Not exactly, but you know what I mean.
Hi all. See VibeKanban.com - it's a Kanban style multi-agent orchestrator that sits atop Claude or Codex. I picked it up on Monday and am overjoyed. It's great.
This is really cool. I've used it for a couple hours today and it might just be the tool I've been needing. I'm not organized enough to be a project manager, and I have a very one-track mind, which makes running multiple branches of a project really hard for me. This helps me avoid forgetting what's going on and allows me to clean stuff up when I'm done. Awesome.
The JV analogy is spot on.
The bigger issue for organizations is that the IT fiefdoms we’ve all grown to know and despise are over and done with - and that means the most supporting those careers is collapsing.
It’s a bit like the power Bitcoin gives to those relying on legacy authority.
There will still be IT careers but the bar to add value is considerably higher and you might need to support a large enterprise to make it worthwhile over just using the tech to build out your own products.
Yes, I have always been an "automate myself out of a job" employee. Which at most orgs is highly valued besides the current one I have bc their internal politics and culture is so shitty. Like I'm a sr devops engineer, and that is a fake job ( full stack developer is fake too), especially in the age of ai.
With AI, every engineer should become a manager of a team of machines doing all the things they're not skilled at. So instead of a team of devs supporting one product, you can have 1 dev supporting an entire product OR a single cross functional team of devs supporting multiple products. Most companies will fail this transition (rightly so) and go the way of the dodo, unless they are conscious and actively doing really&d and retrospectives.
should become a manager of a team of machines doing all the things they're not skilled at
=> should become a manager of a team of machines doing all the things they ARE skilled at
Yeah, both models probably need to be experimented with to see what is most effective. Mature companies probably benefit from your model, startups probably mine until they can get the right people. At least that is my guest.
inb4 you start generating code too
I do, it still only gets about 80% of the way there and i end up rewriting it anyways. Its nice to get the problem started