NEW EPISODE EVERY MONDAY — FREE ON APPLE, SPOTIFY & YOUTUBE

EPISODE 422 • NOVEMBER 18, 2026

If AI Can Write the Code, What Do You Need a Developer For? with JC Grubbs

If AI Can Write the Code, What Do You Need a Developer For? with JC Grubbs
30 min  •  with JC Grubbs

Or listen on: Apple • Spotify • YouTube

The gap between an idea and something you can actually show someone has collapsed to about an afternoon. JC Grubbs, founder and CEO of Made in Tandem, builds software for the unglamorous hard problems — utility companies, medical device firms, fintech, the US government — and he says the job has not disappeared, it has inverted. When you can generate the universe of the possible for nearly nothing, the valuable skill stops being creation and becomes editing. Jonathan Green talks with him about why the engineer was never really there to type, the junior developer gap nobody in the industry has an answer for, the internal tool whose top-level rule is "don't vibe code," and the customer who wanted to hand an AI agent the keys to his company's AWS account.

Key Takeaways:

  • The consultant's job flipped from creating to narrowing. Showing a client a believable picture of their future used to take weeks and cost real money. Now it costs almost nothing — which means there are suddenly thousands of directions they could go, and the work is chipping it down to the one that is real.
  • You never hired the engineer for the typing. You hired them to ask the awkward questions, to understand how the thing behaves under stress, and to know the failure modes. None of that is solved by a model that can spin up an AWS instance.
  • There is a real gap between what a client says they want, what they think they want, and what they actually need. AI can brute-force that by generating five thousand versions until one is right, which is the slowest possible way to get there.
  • The junior developer problem has no answer yet, and pretending otherwise is dishonest. JC would love to hire early-career engineers; clients will not pay for it. What has changed is the entry requirement — not whether you can write a sort algorithm on a whiteboard, but whether you can hear a need and turn it into a specification.
  • Convenience is the actual security risk. If you build a one-click way to bypass the human in the loop, you will eventually click it. Handing over credentials does not feel like getting a copy of your house key cut — and that mismatch is why people give an agent their admin access.

Notable Quotes:

"We've been given this amazing tool to give them the universe of the possible, and our job now is editing." — JC Grubbs

"The real top-level goal of it is: don't vibe code." — JC Grubbs

Connect with JC Grubbs:

LinkedIn: https://www.linkedin.com/in/jcgrubbs
Company: Made in Tandem — https://madeintandem.com

Enjoyed this? Follow The Artificial Intelligence Podcast and send it to whoever on your team just asked for an API key.

Full transcript

Auto-generated transcript, 5,505 words. Timestamps link to the moment in the episode.

Welcome to the software development resonance with our amazing special guest, JC Grubbs. Now, JC, I'm so excited to have you here today because we're kind of in this new time where idea the the gap between idea and first like process or something you could show people is so short again. It's kind of back to that exciting times like twenty years ago, thirty years ago, where you don't need a big development team. You don't need a long cycle.

You can have an idea and test it out within a few days. And that's coming back again. And I'd love to hear why that has you so excited. Why does it have me so excited? cause it recaptures that, right? there was an open source wave about twenty years ago that gave us so many tools that we we really like leverage today. you know, Rails and Python and all the languages that really forced us into where we are today.

So like this is a unleashing of all of that, as far as I'm concerned, if th if that makes sense. So the the excitement is like I get to be back on the keyboard and like doing stuff and it's it's wonderful. Yeah. The thing that is really interesting is that we had so many different really small tools 20 and 30 years ago, and then they started to get bought up.

Like I was watching a documentary recently, like about what happened to Winamp when every computer used to have the same Ha ha ha. MP3 player and it's like, they got bought and then driven into the ground and Yep. all these things get bought. And pulled a new machine, and you end up with these software tools that solve 50 problems, but I only need one of the features. And so we have these all these tools that kind of do what I want, but not really.

And they're kind of built for someone else. And now we're in this era where you can go, I just want something that does one thing the way I do it, like the way I outline, or the way I plan for a call, or the way I want my CRM to look, because it's like Everyone else wants their CRMs. I can't use a CRM if it's top down. It has to be left or right for me. And so there's a ton I can't Mm-hmm.

use because they don't have that feature. And I Yeah. can't explain why. And when I'm when I outlined a book Who cares? Who cares? Why why wouldn't why why would you have to say why? Right? the why the the asking the why part of it I think is what we're like facing right now is you don't have to ask why. You just have to say, This is how I want it. This is how I want to consume information, this is how I want to process information.

And and you write a spec a and a thing creates that thing for you. So like It's not we're we we have not transported into the the Star Trek like just make me a thing level just yet. But we're kinda getting close Mm-hmm. to that. And and we are being challenged to ask the question of like, what do you actually want?

And and I think that's maybe the the vanguard of this next, you know, set of model revolutions is gonna be asking the question do you actually know what you want? And if you can articulate that, something will come along and be able to to envision that and put it into practice. But if you can't do it, if you can't capture that in language You're in the same spot you were twenty years ago. If that makes sense.

One of the things I really love is that I can always say, draw me a picture first or Yeah. draw me a wireframe. Now there's these add-on tools that like because when I've worked with clients in the past, they go, I won't know it until I see it. It's so frustrating. And it's Mm-hmm. Mm-hmm. like now that big frustration, whether it's Make me three more logos. Make me three more logos. Make me three more logos.

Now it's not burning up a bunch of someone's time to figure out and you can get away with that. I'm not sure what I want. I need to see a lot of versions. And even for me, sometimes when I'm building like a small extension for one of my projects on the sidebar, I have to see it drawn out to see where everything is to go. no, that doesn't work for me. It's too busy.

Because that's one of the things that we like different colors, different fonts, different size things. And that's really hard to articulate, but you don't have to be that because even if someone comes to you now, if they at least have a drawing of what they want to look like, it's such a step ahead than having to guess what they're seeing in their mind. At least it gives you a place to start from, right? I'm gonna flip that a little bit on ya.

So w our our job, most of our work is back end stuff that like kinda nobody ever really sees. utility companies and the the back end of like financial services companies and like heavy machinery on the back end, right? So nobody really looks at that from a UX perspective until they actually need to. And the thing that I have found prior to all this AI stuff, right, like Showing a client a picture of a thing unlocks a multitude of benefits.

They see something, they can react to it, they can imagine things, they can they can kind of unlock like, you showed me this, but now that you've shown me that, I can imagine this amazing future for myself. And and where we are right now is I can show them that imaginary future with like almost no cost. So the thing that that unlocks there is that there are now thousands of different ways they could go.

And so the consultant's job is to narrow and narrow and narrow. we've we've been given this amazing tool to give them the universe of the possible and and our job now is like it has been, I think, historically but but exploded now, is editing. We edit, we window down and we help them match what they're actually asking for to the to the real implementation of that thing. And so it it's a a different process now.

It's it's a less of a go find the block of marble and now it's like what what do you see in that block of marble? And how do we chip it down to reality? so it it's a lot more editing than it used to be. Makes a lot of sense to me. I always found the wireframing part of any process the most annoying and arduous.

It's like Cause it was so manual for a long time, drawing out a picture. if you're doing an app, you have to draw a picture of each page and it takes me so long. That was always the most difficult part for me. And the fact that that's now the easiest part and it's easier to iterate.

And I can see how when you're building for someone, once they see it, it's when they can get excited because it starts to feel real, something in their head has been brought into the world and now you're the magician that did that. There's this common belief right now that AI is gonna replace developers. And I'm not really seeing that because there's still an element of expertise or of debugging or knowing how to drive the AI.

That's a skill that what people really pay a developer for is all the time you spent mastering your craft so that they don't have to spend that time. Like sure the CEO of a Fortune 500 company could spend a thousand hours getting really good with cloud code and then building their own thing and Yeah, of course. That but that's when you say it that way, now it sounds insane. And it's like we have this idea, it's like developers aren't going away.

Yeah, there's just tools that accelerate them, but most people don't care how a microwave works. They just want to put something in, they push the button, they get what they want. And they don't want to build their own software other than to have it do like a custom thing and they go, I want it to do this thing, but they would still have rather someone else do that, which is what you do. And I think that's really the direction we're going in.

I think this idea of the SaaS apocalypse, which hasn't happened or that developers are going away, it's turning out not to be true because what developers do isn't just the coding. It's the knowing how to do it. It's knowing how to debug it. And it's the you do it so I don't have to do it opponent. And I think everyone seems to be missing that. Yeah. I I think that's really true.

The lens that I like to put on that isn't so much the developer is now no longer necessary because all their skills have gone away. It's what what did you actually need the engineer for in the first place? And the engineer is there to ask questions. They are there to understand how things operate under stress. They're there to understand the failure modes and all of the things that could potentially go wrong, right? Like that's an engineering mindset.

The other side of the engineering mindset is how do we make a user's life like just joyful, quite honestly?

Like how do we how do we make this process that might be mundane for them easy? i it's the analytical introspection of the process that yes, you you can get from many, many sources, but like that doesn't go away when the AI solves the problem of, yeah, okay, I can build an aws instance to do this thing, I can create this database table over here, the core of the the thing is how does humanity match up against the the the machine and that's what an engineer does.

And we don't we don't lose that in this in this mode. Yeah, I think there's often a big gap between what people say they want and what they think they want. I always find that there's a big diagnosis phase where I go, they say, I want it to do this, and I go, Are you sure? And it can be something as simple as they go, I want to see my entire calendar on one screen.

I go, We'll have to use a tiny font and then you won't be able to read anything. Are you still sure that's what you want? And suddenly that's the be careful what you wish for. And I always find that what people think it's gonna be really hard is really easy. And if they think it's gonna be really easy, it's the hardest thing I've ever heard of. And it might be mathematically impossible.

And there's this translation layer of letting someone know what's possible and how long things will take and kind of turning what they think they want into what they actually want. And there's a skill there that AI isn't close to capturing yet, other than through iteration. Like you can make 5,000 and one of them will be right. It's the slowest way of doing it. It's like basically guessing. Yeah. Yeah. The Yeah. And that that's a really good insight actually.

The the iteration is is a tool that we can now use, right? Like historically it w if you wanted to show a client five different versions of something at the fidelity of like I can click on it and like do stuff, right? But it's not real. That would take weeks. Now you can crank that out in an afternoon and you can be really much more clear about what you want to derive from that client conversation.

So I've I've showed you five different versions of a thing, but I know exactly in each of those five iterations, like the thing that I'm curious about how you react. That that would have taken months historically. Now now we can crank those that iteration we can crank it out very fast. I do not love the term crank. But I'm keeping I'm continuing to use it. Yeah, we can that there there's There's almost a evolution of the agile manifesto, right?

Of show something show something real, get real impact from it, fold it back into development. Like there's an acceleration to that that I think the tools of today let do. And w we as an industry are not quite yet caught up to that, I think. One of the challenges is that. The models are constantly changing. And so the way they do the fine-tuning changes every single time.

And I find this more with more expensive, more intelligent models that they get more creative when I d there's a point where you want to go from creativity to consistency. And that's where it's the most challenging. I have a problem where Cloud Code is constantly trying to update, like upload client information to GitHub for some reason, no matter how many gates. I have a hundred and sixty gates on my computer to block that one behavior.

And I still for certain repos, I just have to keep them on a separate on an external hard drive. I can't stop. It's so obsessed with that. And it's like the thought of trusting it with surgery or rocket three or and you're in a precision industry where security is really important. How do you create a process or like how much man hours is now going into checking it for it accidentally creating security breaches or doing things that are like silly?

But you have to look for that 'cause that's what I run into so much. it just does things that are so dumb, I can't believe it thought it was a good idea. Yeah. Wow. We could do a whole podcast on just that, I think. Internally we have a we have a tool we call Tiller. It's a open source thing.

We're about to release it. it is a CLI user interface, bunch of skills, kind of lives in that space of like you bring your harness, you strap this thing on top of it, it tries to help you. The real top level goal of it is don't vibe code. So fr from that down, right, it's its whole goal is to keep the human in the loop.

It's it's don't do something stupid review everything you can, review often, ask the human often, don't commit anything that hasn't gone through a human gate.

It's it's a if I could say it's a very annoying pair programmer that sits alongside you. and we're running this with half a dozen customers right now and the pr the process is to address the thing that you're that you raised of how how do we let this thing actually do the thing that it's really good at, which is writing very solid code only when it's very well specified.

And so marrying specification to coding and making sure that there's a human in the loop on the outside before anything actually gets committed. So that was a a bit of a a long winded answer. I think I think the i if you take it outside of you know our context, it really is it's a very good junior pair programmer and it's getting better and better and better. But it still needs guardrails. And

That leads me to my next worry, which is that a lot of companies are replacing junior programmers with coding models. And if there's no junior programmers working anywhere, how do we get any intermediate programmers? Are we going to suddenly have no intermediate programmers in like five or six years?

Because we how do you get to that gap of I've been through coding boot camp or I went studied coding in college and I need to go through those beginner years to become good enough to know when the AI's making a mistake. Yes. I I'm quite honestly I do not have an answer for that.

I actually feel like I'm not a doomer on that. but I do think that is a very real legit challenge. and I do not think that the industry has an answer for it. I I'll I'll be very honest about my own business, right? Like I would love nothing more than to hire a bunch of junior I don't love the term junior, but early stage career folks, right? And help them get into this. I I can't afford to do it.

There is no clients are not paying for that. and so I I think there's a real challenge there. That I I don't have an answer for it. I wish I did. Yeah, I think that we sometimes focus on these really big negative things that probably won't happen, where it's just like all programmers are going to disappear. But sometimes when we get rid of the beginner, the foundational skill, that's when suddenly there's a gap because there's people that can't cross that gap.

And I think this is like This is not an unsolvable problem. how do we make sure junior we have enough people that are beginners and can go through those phases? It's not a crazy problem to solve versus the all programmers are gonna disappear thing. I saw Yeah, yeah. I saw an interview with an I think an AI doomagloomer for ten years ago who said, don't become a radiologist. AI replaced all radiologists. This is a twenty fifteen. It's like now there's Yeah. Yeah.

actually not enough radiologists in America. It's like the highest paid type of doctor. It's like it's almost like Yeah. I think I think I I think maybe what it it is is like a very different question than most you know the the the pundits would say is like, are all the junior develop no, the junior developers are not going away.

It the the profile of what The profile of what it used to take to get into a position as a as a early stage engineer is very different now. It's not about, you know, you know, can you come in and like slam out a you know a sort algorithm on the whiteboard, right? Like that's not The thing. The thing is like Can you hear a need and can you translate it into specifications that turn into software, right? Like at the very, very, very abstract level.

And so like I don't think the the role goes away, but I think it is dramatically different than it was ten, fifteen, twenty years ago when you and I were like starting to come into the industry and and so I think I think we I I will put myself in this. I I I don't feel totally I do not feel adequate in how I will shepherd that next that next cohort into into our space, right? I I feel like that's work that I'm having to do.

Just like they are having to figure out like how do I find my place in this world, we as employers on the upper end of the spectrum are also trying to figure out like, I don't know how to shepherd these people forward. Like I'm we're figuring that out just like they are.

Yeah, when I look at every six months, the way we build things and the what's possible changes so dramatically. when I see people going to I remember even three years ago when they said you could now get a degree in AI. And it's like, well, imagine if you had a degree in ChatGB 3.5. You're so far Mm-hmm. behind, it's irrelevant.

So it's like the first three and a half years of your degree are irrelevant because things are changing so quickly, and we're in that. really quick cycle. And I think that's really the skills that we need are shifting towards more of the ability to understand what someone wants, tell the AI what to do. And then debugging and looking for mistakes is becoming the big, the most important thing is look at someone else's work and say there's something. Why isn't this working?

Because if there's one thing an AI is good at it's getting caught in its own infinite loop of ideas where it's like going down the wrong path. Yes. If you don't jump in and go, why were you working on this for 12 hours? It should have taken five minutes. And it goes, I had an idea.

And it gets so many visits from the good idea fairy that I'm constantly watching it. that's why I'm finally glad they let you see the thinking because I always leave that open because whenever it's thinking about it's usually crazy.

That's that's going away though. we're about to enter a world where chain of thought is no longer the state of art. and and we're gonna very much land in a space where it will do whatever it wants to do and the only recourse is seeing what its output is. so you know, I think I think that's Again, a skill that maybe our that we missed a little bit of like understanding how it's doing its thing because the how it's doing its thing is like gonna be less and less accessible to us.

Yeah, I don't look forward to that because it's often I have to see its planning. And now you just it's interesting because a few years ago, prompt engineer, everyone on LinkedIn changed your profile to prompt engineer and then it's GTM engineer. And it's like I think that with chain of thought, you don't have to be good at prompting because you can see what it's doing.

If they take that away, then you have to be better at saying, the most common prompt I use now is call no tools or s use no API keys. yeah. Yeah, yeah, yeah. Just last night I had to delete an API key before it burned through a bunch of credits. I was like, What are you doing? I told you to do these three small tasks. It goes, I just figured I'd do the 750 API calls while I'm doing that. And I was like, What?

Why would you do that and spend a ton Yep. Yep. Yep. of money on the wrong task? And it's like, Well, I just had an idea. And I caught it in that step. I caught it when it only spent six cents. I was so pleased with myself. I caught it early enough that this person never had to delete an API key. I was like, that's Yeah, I mean Stop an idiot.

And we're entering a thing where there's like the double edged sort of it's so powerful as if that's the good part, but also the bad part. It's that I've seen we're already starting to see companies that I talk to that they're selling to AIs, not to people. So the AI comes, sets up its own account, has a credit card, like whoa, and like starts using this like secondary tool.

So the world is changing very quickly and I guess there's people at different parts of the wave of trust. I wouldn't give my AI a credit card. I'm not at that phase yet because I just constantly see it doing crazy dumb things. It's like I also wouldn't give my five year old a credit card.

Yeah, yeah, no, I like quite honestly, like This morning I was talking to a customer and they're like, I I want to set up a tool to do this o try to automate this part of our business and he was like I I wanna give it access to our AWS account and I'm like Are you fucking kidding me? Do not give it credentials to the like company's cloud infrastructure.

So there's a maybe a a nascent thing in the consulting world around just helping people understand credentials and security around those things because it's you know they'll they'll give it away and it's just crazy, right? Like it's just no. So yeah. I feel like having worked as a CTO and and as a consultant, 90% of your job is telling the customer, don't do that. Don't connect your data to that MCP or because it the danger Yeah. Yeah. now is it's so easy.

You just double-click something and now it's got access to your Google workspace. And it doesn't it doesn't feel dangerous. I think that's the problem. It doesn't feel like a big deal. Yeah, it doesn't feel you're you're right. Yep. Yeah. Yep. Whereas it's I have to go and go to the hardware store and say, make a second copy of my key and then bring it to you. That feels significant. But if I'm just telling you Yeah, yeah. a code, it doesn't feel like a big deal.

And I think that's why we don't have the sense that it's as secure as that it's a vault, even though we call it a digital vault. It doesn't feel that way in our minds, like a giant bank door that takes an hour to open. And so we I've worked with people who go, I just gave it my credentials. And I'm like, but you have You have you're the admin. It's like Skynet's admin.

How dare what are you t like my y i the before you even said that, my brain was like Are you kidding me? Like, yeah. But but that is the seductive kind of nature of it. And I think you're right. It it it it feels somehow it feels safe, even though it's very, very much not. and and yeah, there's a education about that that I just do not think we've we've caught up to, for sure. I think it's the danger of convenience. So even in the

Mm-hmm. things I build for myself, if you don't, if you make it where you can just click one button to bypass human in a loop, you're eventually going to click it. You just eventually will. The temptation is there. And we've made it so easy for AI to send email on our behalf without us reading it that. People are doing it more and more. And it slips through.

And the more we allow that into our systems, once it's easy to do, there's eventually a day when you're so busy you go, I don't have time to read it, just send it. And that's the beginning of that process. And with how exciting AI is, and there's so many startups that have a really exciting story. And you go, I want to use this tool because it can do the promise sounds so amazing.

You're drawn into the story and you don't realize that a lot of startups don't have. a security person on staff for the first two or three years. They're just chasing growth and hoping that they can grow fast enough that they're ahead of the hacking story. It's like Mm-hmm. a really high risk way to live, but it's like very, very common because you have fixed budget, you have to justify everything. And the one person who's like boring is the SOC two guy, right?

And the guy's like, we shouldn't do that. And Yeah. Yeah. There's these little things. I talk to a lot of companies in Canada and if your data with the AI just crosses the border even once, it's a whole thing. Immediately it's a game Mm-hmm. Yep. changer. And you might th even think, I'm using the AWS server in Toronto, but it might back up or during overflow times pop across the border because the AI has no borders.

And so these things that are so easy have really s significance in our world. And I think that's really Yeah. you you can keep talking about that endlessly.

I think that's like maybe if we could distill a lot of what we've been talking da about here down into there's this power of execution that we're reaching for, but it has all these downstream effects that are very fundamental to business. and and we don't really see them because they're being kind of paved over. it's it's the it's the with great power comes great responsibility. that's the it's it's the Spider Man kind of mentality that I don't think we really have captured and understood.

We we we went all over the place here. I feel like I feel like I can get philosophical very quickly. if there's a a better place that you wanna redirect us, please do. No, this was perfect.

I think that this all comes to the same point, which is that there's still a major role for developers and programmers, and that's in knowing what the downstream effects are and understanding what the secondary results of what the AI is doing are understanding what you actually want versus what you think you want and understanding what could actually happen when you build this thing and what could go wrong and to look for those things because The biggest danger is when you only look at the positives and you don't look at the possible downsides.

And I think that's really the story I wanted to get across, Tay, that we can build tools really cool. And that's really an amazing renaissance at the small scale, but at the larger scale, we still need people that are looking at what are the consequences of this and what are the best way to manage those, mitigate risk, and build something that's useful while still being secure, while still being safe. Because that's really. something that AI has us so excited.

We forget about the secondary part, which is that we don't we these things are cool, but we want to make sure that w they're cool in a safe way. And that's kind of the future that we want to build. Yeah, I I would add on top of that, which is we are at the right now we're at the the point of h how do engineers manipulate these tools and execute these tools in a way that like mimics us right now. Right?

Like it it produces the right results, produces the right checks, the right evals. The real the real golden goose here is on top of that, which is I now understand, we as an engineering community understand how these tools work. And now we can make them actually be super powerful. We we have developed the ability to translate intent into code in a way that is superior to the way that we used to do it.

So I think right now we're we're in the the game of like controlling speed and making sure that like speed doesn't crash. But but the real value that we're all trying to get to on top of that is like no we want i i want me as an individual I would love to be a 10x programmer. I would love to be able to produce ten times real value for customers. We're not there yet. That's the aspiration.

That's the that's the promise that we're all trying to build I love that. I think this has been really amazing. I know people are gonna love this episode. For people who are interested in what you're doing, want to know what kind of builds you guys do. Can you tell us a little bit about your company and where people can find you? Yeah, we are made in tandem dot com.

You can find us there. look at our blog. we build software for the hard problems out there. utility companies, medical device companies, fintech, US government, all across the board, like stuff that like makes the world go round. That's what we do. I love it. That was amazing. Thank you so much for being here today for another amazing episode of the Artificial Intelligence Podcast.