Announcer: You’re listening to the Call Kent Podcast where Kent C. Dodds answers questions and gives insights to software engineers like you. Let’s hear the call!

Uvindu: Hey Kent! So I’ve been following you for a while and that’s one big question that I have.

So I’ve been working as a front-end developer slash mobile developer for a while. So I mean with the current trajectory of the dev environment and the world with AI is going, I just need to know what is it best to focus on in the future.

Is it something like MCPs and agents or is it something else? I’m a little bit in a dilemma over here. So how should I kind of like continue the learning for me to be improved in the future?

Yeah, that’s pretty much it.

Announcer: And that was the call. Here’s what Kent had to say.

Kent: Hey, thanks for the question. Hey, thanks for the question. I think your name is pronounced Uvindo. So thank you so much for asking.

This is actually something that I’ve been thinking about a lot. And I’ve talked with a lot of my friends who are software engineers as well. And we’re all having occasional existential crises around the way that things are changing so much. And it is like determining what the trajectory of the world is. It is kind of a difficult thing to determine. And honestly, it’s impossible. So I’ll share a bunch of thoughts that I’ve had recently.

So first of all, when Opus 4.5 and GPT 5.2 came out back in like, I think, November, that changed a lot. They were a massive improvement in agentic development. And now I very rarely have to manually review code. I’m certainly not writing the code. I am prompting and my prompts can be relatively simple.

For example, before answering your question here in the Call Kent podcast, I decided it would be nice if the caller transcript was generated before I asked the question. So before, a few minutes ago, when I received a call, I would respond to it like I am now. And then our call would be stitched together and we’d make a transcript and it’d go up to the podcast host.

But I realized, oh, it would be actually really nice for me to be able to look at a transcript visually even while I’m answering the question. So how about we have the colors portion of the transcript be generated ahead of time?

And so I made a prompt and I can pull it up right now because I literally just did it a few minutes ago. My prompt was one, two, three sentences long. I’ll read it to you.

It was really helpful in the Call Kent podcast or it would be really helpful in the Call Kent podcast if I could see the colors transcript. Right now we generate the transcript when we get the episode ready to go. I would like to generate the colors transcript portion before that so that I can see it before I record my portion.

The agent worked for 18 minutes. This is all in cursor cloud agents. And then it showed me a video walkthrough of a test showing that that transcript being generated.

Now, one of the things that makes it so successful is that I was able to close the agent loop so the agent could actually run this test. I don’t give it my API keys to actually go through the endpoints to make a transcription of stuff, but I have everything mocked out so that I can.

And so what I’m seeing in the walkthrough is the mocked version or the mock response for the transcript, which is perfect. That’s that as long as it’s calling the mock properly and everything, then I can assume that it’s going to work in production because I’ve already tested that myself.

And so I was happy with that. But then I had the idea that it would actually be nice for me to to change something. So here what I said is, can you make it so that I can edit the I did this as voice to text and it says edit the color transcript.

That’s not what I meant. That’s not what I said, but that was my voice to text interpreted what I said as color transcript, but edit the color transcript. And also that we use the call transcript instead of generating it when we’re creating the transcript for the whole episode.

So I noticed I actually I think I looked at the code briefly because it finished it so quickly and I was surprised. So I did look at the code briefly and I was pretty sure that it wasn’t actually that it was regenerating the transcript of the color when it was generating the full transcript for the episode.

And so I just said, hey, make sure that we’re using the part of the transcript for the color that we already generated. And also, I would like to edit that.

And so it worked for another 37 minutes here in the cloud. And then it made me another video to demonstrate that it was working. And that’s what I’m looking at right now.

And then the pull request is open and it was ready to go. And bug bought the cursor review tool went through and iterated on a couple of things. Like one thing that it fixed was fire and forget may cause stale status on redirect.

And there was another one that, yeah, text area won’t display generated transcript after pulling completes. So it’s finding all these edge cases.

But then I also have code rabbit on this. It’s an open source repo and so it’s free. And so code rabbit had a couple of things, too.

And so the agent just iterated with these bots to catch edge cases and different things. And then I had a rebase on merge because I or on main because I made some changes in another branch and that was it.

So like I wrote none of that code. I barely looked at any of it. And I think that this is only going to get better.

Like six months ago, this would not have been a thing. You would not have been able to do this. This isn’t like a non trivial change. It involved a database migration and a bunch of stuff.

I did this while I was doing other stuff. I probably spent a good three or four minutes working on this feature. And now I’m literally using it right now.

And the transcript for your portion of the call was actually almost perfect. The only thing I had to change was when you said front end developer slash mobile developer. It used the word slash and I just changed that to be the character slash.

Other than that, it was a perfect transcription, which is crazy to me.

So anyway, just like what my point is in this whole thing is that things have changed drastically. You could not have done that. What I just did could not have done that six months ago or even four months ago.

The models have gotten so much better and the tools that use the models have also gotten so much better. And so when I saw like how much improved everything was back in like January, I had an existential crisis myself.

Because OK, so my job for the last 10 years has been to take to help experienced software developers acquire or accelerate their acquisition of experience in a domain that they have no experience in yet.

So you know how to code, but you don’t know react yet. Well, epic react will get you there. You know how to code, but you don’t know testing yet while testing javascript dot com is going to get you there. And the same for full stack development with remix and and MCP and all of that.

And what I was coming to realize is that agents are making it. So any experienced software developer doesn’t need to have experience in a specific domain anymore because the agent can fill in those gaps for them.

And so that was kind of what my my thinking was back in like January and I was freaking out and I ended up going or starting the interview process at a couple of companies. Decided not to go with any of them just because I I don’t think that I would be very happy working for somebody.

And I really, really feel like I can still make a living doing what I’m doing.

And so anyway, to fast forward a little bit, I launched the TypeScript foundations to fluency, the practical TypeScript course on Epic Web that actually incidentally is still on a 40 percent discount at the time of this recording.

So if you’re looking for a stronger foundation in TypeScript, go for it.

But yeah, I and actually I do think that having some experience with coding is valuable with actually coding is valuable for using these agents effectively. So if you’ve got code curious friends, then definitely recommend it to them anyway.

So I launched that and now I’m like, OK, what’s next? What am I going to do?

And my thought was, OK, I’ll teach workflows. But here’s the problem. Six months ago, my workflow was completely different. In six months, I’m pretty sure that my workflow is going to change drastically again.

I don’t know what it will change to. Six months ago, I didn’t expect that I would be able to do what I do now.

And so the problem with doing like, let me teach you how I develop websites in 2026. That was that was my plan. The problem with that is twofold.

One, the developers who are learning are going to learn something that will then be no longer valuable in six months. I mean, maybe I’m adding I’m making stepping stones so that they can step to where we are now. And then it makes it easier to step to where we go next or something. So there’s some value there.

But I don’t like the idea that the knowledge that I’m giving somebody is no longer relevant or falls out of date so quickly. I don’t like that.

The other aspect to that is the way that I’ve been able to scale my business so that it sustains me and my family long term is making it so it’s self paced.

And there’s a lot of benefits to self paced as well. So like some people can’t make it because the time zones are bad. Like you can’t make it to a live workshop.

And also, I have to charge more because it’s my personal time. I can’t do purchase power parity and stuff like that. It just makes it really hard for me to make it sustainable.

So making something that’s self paced solves all those problems.

The problem with that, though, is that it takes a lot of work to make something that’s self paced. And if and the what makes it worth it is if that self paced thing can continue to make me money and provide value in the world for years in the future.

Well, if I do something that’s teaching you workflows, that’s not going to work. Not at all, because it’s like it’ll take me three months to get it released. By the time it’s released, things have changed so much already.

So it kind of defeats the whole thing.

And so then my thought was, well, OK, maybe I just do live workshops all the time and we just do the very best that we can with that. But again, that doesn’t solve the problem where things are going to change again in six months or even three.

So I just decided, OK, well, everybody’s worried about AGI.

And if if AGI happens like artificial general intelligence, if that really happens, then we’re going to be in a really different place. If things are going to be like everything changes in that environment.

And that’s definitely a problem for just everybody and not something that you can really plan for. I have no idea what the world is going to look like in that place.

So because you can’t plan for it, we just take it off the table. We’re not going to expect that’s going to happen.

So if it doesn’t happen, then programmers are still going to be programming in 10 years.

And if programmers are still programming in 10 years or five years, then what like what are the things that they’re going to need to know?

That’s very difficult to understand, but or to like divine to know.

But I think that bringing clarity to problems. That is what we software developers will still need to be able to do.

We needed to do that 20, 30, 50 years ago. We need to do it today. I think we’ll need to do it in the future.

Bringing clarity to problems, figuring out what the problem actually is and how software can be used to solve that problem.

That’s what we software developers have been doing since the beginning and what we will continue to be doing in the future.

And so that is what I am planning on teaching people is how to bring clarity to problems and then take that clarity and hand it over to an agent to build the solution and then iterate with the agent and with stakeholders to make that solution what they need.

And so that means I’m not teaching the workflows. You will incidentally learn workflows as you go through this material, because that’s just kind of how you get to the solution.

But really what I’m teaching is how to communicate with stakeholders and end users to build a tasteful solution to their problems.

So this is a pretty significant experiment for me.

Actually next Wednesday, I’ve got a group of friends that are going to be joining me for a trial run of this workshop so that I can kind of figure out how does that change things.

It still is using the Epic workshop app, but I’m going to need to make some significant changes to that to facilitate this type of workshop or something.

So yeah, it’s a big experiment. I’m not sure, but I’m feeling jazzed. I’m actually really excited about this. I think it’s going to be great.

So anyway, with all of that said, what should you do? Should you be working on MCPs or agents or something else?

I think that like continue to work at the company that you work at and solve the user’s problems.

Figure out how to convert the conversations you’re having with stakeholders into clarity for the problem space, and then hand that off to the agent to solve and iterate with the agent.

And I do think that the days of our coding by hand are already over for me. Like 100% of my code is written by agents now.

And being able to do that and have good workflows is still important for that. But those things are going to constantly be changing.

I would not spend a bunch of time fine tuning a workflow because it’s going to change soon anyway.

And yeah, I guess the like as far as the specific problems and solutions and things, I do think that MCP is going to be huge. I still feel that way.

It continues to become more and more relevant, especially in the enterprise, but for consumers as well. That’s that’s where I’m most interested is for consumer. I think that’s going to be huge.

Custom agents, I think, will be relevant for sure.

And in fact, kind of putting your everything that your system does behind an agent and then exposing that agent as an MCP server is something that I think is going to happen.

So understanding how agents work.

In fact, I actually wrote a blog post a couple months ago about what I think you’re going to need to know in 2026. So I’ll link to that blog post in here as well.

In fact, hold on one second.

All right. Yeah, I just paused it and I’ve got the blog post open.

So I’ll read off the like highlights and then you can take a look at the blog post later.

So the first first one I have on here is to keep building software.

Then stick to proven software principles. You can take a look at the epicweb.dev slash principles. I do think that this is really important.

Number three is practice context engineering, which honestly, like this is mostly around practicing workflows and stuff. At the time that I wrote this, my workflow was completely different.

So I would de-emphasize practicing context engineering and mostly just use the best, use a tool, find a workflow and use that and get pretty good at it. But don’t spend a ton of time fine tuning that workflow because it’s going to change.

Number four here is build MCP servers. I do believe that for sure.

Number five is build an AI agent. I’ve got some good thoughts on that. Definitely read the blog post.

Six is build vectorized search. I actually built that for my own website recently. And it’s, it’s very, it’s very awesome. To be honest, it’s yeah, super, super cool.

So that would be something like vectorized databases and understanding that I think is very helpful in this natural language semantic meaning stuff that we’re doing now.

So definitely do that.

And Cloudflare has some really good resources for that. I wrote a blog post about how I did it for my site. So take a look at that. I’ll add a link in the description here.

And then building evals, AI testing and observability.

That’s one area that I, I haven’t dove into myself. I think I have mixed feelings about evals because it’s like the kind of test that actually costs you money.

But if you don’t have reliable like if you can’t make reliable changes to your agents, then maybe that costs you money too.

So yeah, that one’s, that was kind of an interesting, interesting one, but that’s the last one. I’ll add a link to this in the notes here.

But I definitely, I get you. And I think a lot of people get you and they’re just kind of worried about what to learn.

What should we be focusing our time on? How do we prepare for the future?

And I hope some of my thoughts were helpful to you as you’re trying to figure out what you want to do.

With that, I’ll just say thank you very much for taking the time to call in. And I wish you the very best of luck in your pursuits.

I hope that I am able to be helpful to you with this new workshop that I’m planning on.

And it’s going to be a series for sure.

But yeah, that’s my hope for 2026. Stay tuned.

Announcer: This has been the Call Kent Podcast. Learn more about Kent at KentCDodds.com and get your own questions answered at KentCDodds.com/calls.