2026 Predictions: Will We Still Write Code, or Just Manage Agents?

Summary

This video explores bold predictions about the future of software development in 2026, featuring insights from researchers and leaders at Google DeepMind, Cursor, Replit, Sourcegraph, MiniMax, and others. The central thesis is that coding as we know it is dissolving - developers will transition from writing code themselves to managing fleets of AI agents that do the work for them.

The most provocative prediction comes from Steve Yegge at Sourcegraph, who claims that by summer 2026, traditional IDEs will be obsolete and anyone still using them “won’t be a good engineer anymore.” Instead, developers will use agent orchestrators - consoles where they check in each morning to see how their agents performed overnight. Multiple speakers emphasize that 2026 will be the “year of non-technical people coding” as millions gain proficiency with coding agents.

Beyond coding agents, the video covers predictions about multimodality (combining video, image, and code generation), voice agents, continual learning for AI models, robotics assistants for home use, and the need for better enterprise adoption of coding agents for complex legacy codebases in Java, C++, and C#.

Highlights

”If You’re Still Using an IDE by Next Summer, You’re Not a Good Engineer”

yt-dlp --download-sections "*1:11-2:15" "https://www.youtube.com/watch?v=qqvODjOezX4" --force-keyframes-at-cuts --merge-output-format mp4 -o "qqvODjOezX4-1m11s.mp4"

“My hot take is if you’re still using an IDE by next summer, you’re not a good engineer anymore.” — Steve Yegge, 1:52

”How Did My Agents Do Last Night?”

yt-dlp --download-sections "*1:35-2:30" "https://www.youtube.com/watch?v=qqvODjOezX4" --force-keyframes-at-cuts --merge-output-format mp4 -o "qqvODjOezX4-1m35s.mp4"

“You’ll come in in the morning with your coffee and you’ll be like, ‘How did my agents do last night?’ And it’ll show you this one’s stuck, this one’s finished, this one needs your help.” — Steve Yegge, 1:35

”Agent Users Will Become More Managers Than ICs”

yt-dlp --download-sections "*3:51-4:50" "https://www.youtube.com/watch?v=qqvODjOezX4" --force-keyframes-at-cuts --merge-output-format mp4 -o "qqvODjOezX4-3m51s.mp4"

“I think 2026 agent users will become more managers than ICs.” — Michele Catasta, 3:51

”Millions of Non-Technical Users Creating Advanced Software”

yt-dlp --download-sections "*4:22-5:20" "https://www.youtube.com/watch?v=qqvODjOezX4" --force-keyframes-at-cuts --merge-output-format mp4 -o "qqvODjOezX4-4m22s.mp4"

“I expect millions of users, non-technical users, to become proficient using coding agents… and I expect most of them to create something far more advanced than just prototypes.” — Michele Catasta, 4:34

Key Points

• Coding as AI’s ultimate test (0:00) - Unlike poetry or conversation, code is pure logic that doesn’t tolerate mistakes, proving AI can plan, execute, and reason
• Definition of coding is dissolving (0:31) - Generative AI is fundamentally changing what developers do and how software is built
• Developer job changing forever (0:55) - Moving from writing code ourselves to managing fleets of agents that do it for us
• 2026: Year we lose the IDE (1:11) - Steve Yegge predicts nobody will use traditional IDEs anymore, replaced by agent orchestrators
• Morning agent check-ins (1:35) - Developers will start their day reviewing which agents are stuck, finished, or need help
• Hot take on IDE users (1:52) - If you’re still using an IDE by next summer, you’re “not a good engineer anymore”
• Learn coding agents now (2:08) - Recommendation to learn Cloud Code, Codex, AMP, Cursor, or Kline immediately
• Year of non-technical coders (2:20) - 2026 will be when non-technical people start coding en masse
• Enterprise adoption still early (2:27) - Enterprise companies haven’t adopted coding agents at scale for complex Java/C++/C# codebases
• Continual learning needed (2:51) - Models need to learn over time like real colleagues instead of wiping memory clean
• Decade of agents (3:17) - This isn’t just 2025’s trend; it’s a decade-long transformation
• Sub-agent architectures coming (3:37) - Expect products that let you orchestrate multiple agents in parallel
• Agent users become managers (3:51) - 2026 agent users will be more like managers than individual contributors
• Robotics excitement (4:01) - Home assistants performing basic tasks will start appearing
• Millions of non-technical users (4:34) - Expect millions of non-technical users to become proficient with coding agents
• Year of understanding (4:51) - Advocacy for broader developer audience to learn about these tools
• Year of multimodality (5:15) - Gemini 3 and multimodal agents combining video, music, and code generation
• Fast vs smart models (6:00) - Excitement for both fast flash models and smart long-running thinking models
• Year of voice agents (6:29) - Voice is technically challenging but may break through in 2026
• No slop AI coding (6:46) - Hope for higher quality AI-generated code without “slop”

Mentions

Companies

• Google DeepMind (0:49) - Mentioned as one of the organizations whose researchers were interviewed
• Cursor (0:53) - AI coding tool company, featured in interviews
• MiniMax (0:53) - Lab working on video generation, music generation, and multimodal AI
• Sourcegraph (1:11) - Steve Yegge’s company, building agent orchestration tools
• Replit (4:21) - Company focused on making millions of non-technical users proficient with coding agents
• Google Antigravity (5:37) - Introducing new patterns for computer use and image generation with code generation

Products & Technologies

• Cloud Code (1:29) - Command line coding agent, mentioned as current state that will be replaced
• Codex (2:11) - Coding agent tool recommended for developers to learn
• AMP (2:11) - Coding agent tool mentioned as option to learn
• Kline (2:12) - Coding agent tool mentioned as option to learn
• Rapid Agent (3:43) - Next version will allow running several agents in parallel
• Gemini 3 (5:20) - Google’s upcoming multimodal model

People

• Steve Yegge (1:11) - Software Engineer at Sourcegraph, makes bold prediction about IDEs dying
• Karpathy (2:27) - Referenced for his “decade of agents” concept
• Michele Catasta (3:17) - President & Head of AI at Replit, discusses decade of agents and robotics
• Dex Horthy (2:27) - Founder at HumanLayer, discusses enterprise adoption
• Lee Robinson (4:51) - Developer Education at Cursor, discusses year of understanding
• Olive Song (5:15) - LLM RL Engineer at MiniMax, discusses multimodality
• Kevin Hou (5:37) - AI Code at Google Antigravity, discusses multimodal patterns and model diversity
• Anthony Giuliano (6:29) - DevRel Engineer at Block, discusses voice agents and “no slop” coding

Surprising Quotes

“My hot take is if you’re still using an IDE by next summer, you’re not a good engineer anymore.” — Steve Yegge, 1:52

“You’ll come in in the morning with your coffee and you’ll be like, ‘How did my agents do last night?’ And it’ll show you this one’s stuck, this one’s finished, this one needs your help.” — Steve Yegge, 1:35

“I think 2026 agent users will become more managers than ICs.” — Michele Catasta, 3:51

“I expect millions of users, non-technical users, to become proficient using coding agents… and I expect most of them to create something far more advanced than just prototypes.” — Michele Catasta, 4:34

“No slop AI coding 2026 doesn’t have a good ring to it.” — Anthony Giuliano, 6:50

Transcript

0:00 Coding has always been the ultimate test for AI. Why? Well, because unlike a poem or a conversation, the code does not tolerate mistakes. It’s pure logic. And if an AI makes a wrong move, the whole thing breaks. So that’s why coding has always been this milestone that matters. It actually proves that the machine can plan and execute and reason and at the end actually solve problems.

0:29 But now thanks to generative AI the whole field and the whole definition of coding is dissolving. What a developer will look like in 2026. What coding will look like in 2026. Even a year ago the things that are going to happen in the next few months were impossible to imagine.

0:47 I talked to the researchers and leaders from Google DeepMind, Cursor, MiniMax and others and they say that the job of a developer is about to change forever. And we’re moving away from writing code ourselves to managing a fleet of agents that do it for us. Here is how it’s going to happen.

1:11 I think that 2026 is the year that we lose the IDE, that nobody’s going to be using them anymore. By summertime, we’re going to be switching to agent orchestrators, which are a lot like an IDE in that they’re integrated and it’s your console where you see everything, which in my opinion is a lot better than what we have today with Cloud Code, right? Nobody likes the command line interface, right? I mean, it’s just not it.

1:35 And then that thing is going to be - you’ll come in in the morning with your coffee and you’ll be like, “How did my agents do last night?” And it’ll show you this one’s stuck, this one’s finished, this one needs your help. Right? That’s the way it’s going to be. And I think it’ll happen next year. I see companies building it already. I’m building one. A bold prediction.

1:49 All right. Look, my hot take is if you’re still using an IDE by next summer, you’re not a good engineer anymore. And that’s going to be really, really, really, really hard for most of the world’s engineers to swallow. It’s going to be tough. So, I feel bad for them. I feel bad. This is the reality. No more IDEs.

2:08 So, go learn coding agents. Learn Cloud Code or Codex or AMP or Cursor or Kline or whatever. Okay? But make sure it’s the agent, okay? Get ready for next year. The tools will get better, but you can start building those skills now.

2:20 Next year will absolutely be the year of non-technical people coding. It’s already happening in a big way. Next year, it’s going to be just everyone. You know, I like what Karpathy has talked about around the decade of agents.

2:29 It still feels incredibly early in terms of like actual enterprise companies adopting coding agents at scale and deploying it for gnarly code bases. You know, enterprise code bases, you know, Java, C++, C#, not just the in the distribution things like JavaScript, TypeScript, Python, which there’s plenty of. I think 2026, we’ll see models get much better at that.

2:51 I would love, I’m not an ML researcher, but I would love if we started to make more progress on more continual learning of models. Right now, you know, it’s like you wipe the model’s memory clean every single time. More ways that we can help the models learn over time like real colleagues, I think, will be amazing. We can bridge the gap with markdown files, you know, we can bridge the gap with context engineering, but more of that that can be built into the model layer. I would love to see - I know it takes some breakthroughs in ML, but hopefully we’ll get there.

3:17 It’s still going to be the era of agents to be honest. I know that I’m biased because that’s exactly what we’re going to be building, but I don’t think it has been only 2025. I think it’s going to be the decade of agents. We just barely started to see autonomous workloads running on coding agents and they’re far from being perfect for several different reasons.

3:37 So I expect architectures with more sub agents. I expect products that allow you to orchestrate several agents yourself. So like even the next version of Rapid Agent you will be able to run several of them in parallel, see the results that you get, cherry pick the one that you prefer.

3:51 I think 2026 agent users will become more managers than ICs. I’m going to give you two predictions, one about agents and one about AI in general. The one about AI in general, I’m very excited about anything regarding robotics. I do think we’re going to start to see some assistance at home performing basic tasks and there are a lot of announcements, you know, videos that we have been seeing in the last couple of months. Of course, some of them are hype. There is like a ground truth there. There’s something is already working well enough and I can’t wait to buy one for myself.

4:22 I bet on the coding agent side in a sense is a boring prediction but if you extrapolate from how much more powerful they have become in the last 10 months, imagine where we’re going to be you know at the same conference next year. I expect millions of users, non-technical users, to become proficient using coding agents. That’s exactly the company mission that we have and I expect most of them to create something far more advanced than just prototypes. That’s what I promise to our users in the next year.

4:51 I think it’ll be a year of understanding. So for me, I work in the education space and it’s been really cool at this conference this week to meet all of these really intelligent and smart people who are working on really interesting problems. And I think that we’re all sitting at the bleeding edge of this and I’m excited to advocate for it to the broader developer audience and have them all learn about these things and improve their own software.

5:15 Definitely a year of multimodality because we see Gemini 3 and also for us because we are a lab that has like video generation, music generation, all the modalities. So I think that’s where we kind of move to like how humans really interact with the world and you know can you do agentic tasks that are multimodal - that’s definitely one of the themes.

5:37 Antigravity introduces a lot of new patterns that are I would say new - computer use relatively new, image gen in coordination with codegen is also relatively new. So the multimodality of things is - I hope, I believe, or I have conviction that this is an avenue that more and more people are going to hop onto. So I’m excited to see what people do with this in 2026.

5:56 I’m also excited for - I mean models are just going to get better, right? So they’re going to be really good models, but then I’m also excited for really fast models. The models have different characteristics, right? And so you can imagine like a flash model being very good. Introduces different form factors than a very very smart long-running thinking model. And so I’m excited for what 2026 has in store there.

6:14 And then there’s other stuff like world models. So cool. Would love to actually use them in like kind of my consumer daily life. Haven’t found a use yet, but that’s been pretty awesome. I want to try like some AR stuff. I haven’t yet. So I’m a cyclist. It’d be really cool to have more cyclist like wearable type things. So yeah, a lot of things to look forward to.

6:29 I think 2026 might be the year of voice agents. Voice is really really hard. I don’t spend a lot of time in voice, but I know that from a technology standpoint, from an engineering standpoint, from doing like asynchronous programming and optimization, it’s really, really hard. So I think it would be cool if 2026 was the year of voice agents.

6:46 Of course, 2026 being the year of no slop would be great. I would love to see that. No slop AI coding 2026 doesn’t have a good ring to it. So those are my bets.

6:54 Thank you for watching. That was Attention Span and the predictions about coding in 2026. I’m very interested in your opinions. Please leave a comment. Also, subscribe to the channel. Share it with your friends, peers, and even enemies. Thank you so much. I wish you very very happy holidays. And we’ll see you soon.