The argument in one line.
AI coding is shifting from isolated, manually-managed workflows in IDEs to portable, autonomous agents that operate across devices and environments with auto-compaction, queued tasks, delegated permissions, and multi-agent orchestration.
Read if. Skip if.
READ IF YOU ARE…
- A developer actively using AI coding tools (Claude, Cursor, Replit) who wants to understand how agent-based workflows differ from current IDE-native approaches.
- Someone building or shipping products with AI agents in 2025-2026 who needs clarity on which structural shifts are hype versus actionable today.
- A technical founder or engineering lead evaluating how to position your team's development practices ahead of industry shifts in AI-assisted coding.
- A developer comfortable with vibe coding or prompt-based development who's ready to move toward intentional, multi-environment agent orchestration.
SKIP IF…
- You write primarily in non-AI-native stacks (legacy codebases, heavily constrained environments) where agent portability and multi-environment access don't apply.
- You're still evaluating whether to adopt AI coding at all — this covers advanced structural patterns, not foundational onboarding.
The full version, fast.
AI-assisted coding in 2026 is reorganizing around five structural shifts that change how you work day to day, not just which tool you open. Agents now run anywhere � phone, terminal, cloud � with portable sessions you can hand between devices; context auto-compacts so you stop babysitting the token bar; prompts stack into a queue the agent works through while you sleep; permission gates collapse into auto-approve modes because reasoning quality and tool-call volume both went up; and a single operator drives multiple agents across multiple projects, often on non-code tasks like video, research, and assets. The practical move is to lower permission friction, queue work in tools like Codex desktop, and treat your repo as a context folder for an orchestrated fleet rather than a solo editor session.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 01:00
01 · Cold open and promise
FUD hook: vibe coding changed more in recent weeks than most realize. Promise: 5 shifts, the future of AI building, how to apply them today to be ahead of 99%.
01:00 – 03:17
02 · Shift 1 - Agents code anywhere
Moving from AI inside an IDE to agents accessible from phone, terminal, any environment. Tools: Claude Code remote control, Cursor Cloud Agents, Replit and Lovable mobile apps. Sessions transfer across devices.
03:17 – 05:09
03 · Shift 2 - Auto-compaction
Manual context management is becoming obsolete. Frontier models ship auto-compaction that preserves critical context without context rot. Amp leads the way.
05:09 – 07:54
04 · Shift 3 - Stack the queue
As agent quality rises, stop send-wait-review loops. Queue multiple prompts; automate daily code review at EOD. Codex desktop shows this: stack messages, drag to reorder, interrupt mid-chain.
07:54 – 09:15
05 · Shift 4 - Let the agent cook
Power users already run dangerously-skip-permissions 100% of the time. Sub-agents generate so many tool calls that manual approvals are pure friction. Anthropic and OpenAI ship auto-review modes for this reason.
09:15 – 11:09
06 · Shift 5 - Multi-agent multi-project
Run multiple agents across multiple projects simultaneously. Orchestrate so they do not conflict. Non-code tasks (research, launch videos, social, presentations) belong in the agent queue too.
11:09 – 12:03
07 · CTA and outro
Community pitch: school.com/aiapps, 250+ members.
Lines worth screenshotting.
- AI coding agents are moving from editor-bound sessions to portable sessions that transfer between phone, terminal, and cloud as the developer moves through their day.
- Auto-compaction will eliminate the current practice of manually managing context windows — agents will maintain the critical parts of long conversations without the user intervening.
- Context rot — where compaction on top of compaction strips out critical information over a long session — is the problem auto-compaction needs to solve, not just token overflow.
- Queue-stacking replaces the send-wait-review loop: as agents become more reliable, you can stack multiple prompts as a backlog and review the results in bulk.
- Permission delegation lets you authorize agents to complete specific classes of tasks without manual approval on each one, increasing throughput on well-defined work.
- Multi-agent orchestration — one agent calling and coordinating others — is the structural shift that makes complex cross-system tasks achievable without human hand-offs.
- AMP rebuilt its coding agent from scratch based on first principles rather than iterating on its previous product, which is a signal of how discontinuous the current shift is.
- Replit and Lovable mobile apps already let you continue a coding session on your phone, proving that the portable-session model works for users who build away from their desk.
- Rigorous automated code review running in parallel with the build queue lets you benefit from queue-stacking without sacrificing the code quality that manual review used to provide.
- The future of vibe coding is not about writing fewer prompts but about writing prompts that set agents in motion on longer unsupervised runs with structured checkpoints.
- Most AI coding tools in 2026 are still optimized for the send-wait-review pattern — the developers who switch to queue-stacking and async review gain an immediate throughput advantage.
- The five shifts — portability, auto-compaction, queue-stacking, permission delegation, and multi-agent orchestration — are happening simultaneously, not sequentially, which compresses the adoption window.
Five Structural Shifts Define How Serious Builders Use AI Agents in 2026
AI coding workflows
Chris Ashby's five-shift framework shows that the gap between casual AI tool users and productive builders is not model quality — it is workflow architecture: how you queue work, manage permissions, and orchestrate multiple agents across multiple projects simultaneously.
02Shift 1 - Agents code anywhere
- Moving from AI inside an IDE to agents accessible from any environment — phone, terminal, other apps — is the portability shift
- Session transfer across devices means work continues without re-establishing context; the session is the unit, not the machine
03Shift 2 - Auto-compaction
- Manual context management is becoming obsolete as frontier models ship auto-compaction that preserves critical context without rot
- The skill shifts from managing the context window to designing tasks that compound cleanly across sessions
04Shift 3 - Stack the queue
- Queue multiple prompts, drag to reorder, interrupt mid-chain — the send-wait-review loop is the bottleneck to eliminate
- Automating end-of-day code review means the agent is working while you are not — parallelism across time, not just across tasks
05Shift 4 - Let the agent cook
- Sub-agents generate so many tool calls that manual approval is pure friction — auto-review modes exist to remove that bottleneck for trusted sessions
- Permission delegation is not recklessness; it is the prerequisite for overnight and batch agent runs
06Shift 5 - Multi-agent multi-project
- Run multiple agents across multiple projects simultaneously with orchestration to prevent conflicts
- Non-code tasks — research, social content, presentations, launch materials — belong in the agent queue alongside code tasks
Terms worth knowing.
- Vibe coding
- A development approach where the user describes what they want in plain language and an AI coding agent writes all the code, with the human reviewing and steering rather than typing code manually.
- AI coding agent
- An autonomous AI system that can read a codebase, plan implementation steps, write code, fix bugs, and run tests — operating with minimal human intervention beyond high-level direction.
- IDE (Integrated Development Environment)
- A software application that combines a code editor, debugger, and build tools in one interface — examples include VS Code and Cursor, commonly used as the host environment for AI coding tools.
- AMP (coding agent)
- An AI coding agent (by the company behind Claude) rebuilt from first principles to support portable sessions, auto-compaction, and multi-environment access as their vision of the future of AI-assisted development.
- Context rot
- The degradation in output quality that occurs when an AI agent's conversation history is compacted repeatedly over a long session, gradually losing critical information needed to maintain coherent understanding of the project.
- Auto-compaction
- An AI agent feature that automatically summarizes and compresses older conversation history to free up context window space while preserving the most important information — preventing the user from having to manually manage context.
- Queue stacking
- A workflow pattern where multiple follow-up prompts or tasks are submitted to an AI coding agent in sequence before it finishes the first, allowing it to work through a backlog autonomously without waiting for human review after each step.
- Permission delegation (agent)
- A setting or mode in an AI coding tool where the user grants the agent permission to take most actions automatically — reading files, running commands, making edits — without prompting for individual approvals.
- Multi-agent orchestration
- Running multiple AI coding agents simultaneously across different projects or different types of tasks (writing code, creating assets, doing research), coordinated so they do not conflict with each other's work.
- Tool calls
- The individual actions an AI agent performs during its reasoning process — such as reading a file, running a command, or querying an API — each of which may trigger a permission prompt in a supervised environment.
- Context file (agent)
- A markdown file in a project folder that provides an AI agent with background information about the project — goals, architecture, conventions — so the agent can work effectively without re-reading the entire codebase from scratch.
Things they pointed at.
02:14toolClaude Code remote control
02:24toolCursor Cloud Agents
04:20toolAmp coding agent
07:00toolOpenAI Codex desktop app
07:54channelPeter Steinberger
07:56channelRiley Brown
Lines you could clip.
00:18
“You can be ahead of 99% of other people.”
08:44
“You can just sit there hitting yes yes yes yes approve constantly until the AI agent actually completes its work and 99% of those approvals actually did not need your input at all.”
09:13
“Part of being able to do this is actually just a skill issue in terms of how you use these AI coding tools.”
07:54
“There are people who are already doing this months ago using dangerously skipped permissions pretty much 100% of the time.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphor
00:00Wipe coding has changed a lot in the last year. But in the last few weeks, it's changed more than some of you might realize. And most people don't know that this is happening.
00:08So in this video, I wanna break down the five big shifts that are happening in vibe coding today so that you know exactly how to build with AI the right way going into the rest of 2026. So in this video, we're gonna break down the five shifts that are happening in vibe coding right now, what the future of building with AI looks like, and how to actually apply these lessons to your vibe coding or AI coding workflows today so that you can be ahead of 99% of other people.
00:35If you don't know me, my name is Chris. And for the last fifteen years, I've been designing apps and advising startups and products and design. If you are interested in building real applications and software with AI and you wanna learn how to do that and launch to real customers, I've got a community helping people do that and you can find out more by clicking the link in the description down below.
00:53Otherwise, let's jump in and break down the five big shifts happening in coding. And number five is gonna completely change the way that you build. So the first big shift that is happening in vibe coding today is moving from AI coding inside of an editor or an IDE to AI agents that are coding anywhere.
01:10And what I mean by this is being able to access your coding sessions through multiple different agents from anywhere that you are, whether you're out and about and you're accessing your agents on your phone or whether you're sitting at your desk and using your laptop. And not only that, but being able to use agents inside of different environments as well.
01:28So not just inside of an IDE or a coding environment, but being able to access your agents from inside of terminals, inside of other apps, and to communicate between different apps as well. Now most of these big shifts are taken from the recent release of AMP's new coding agent and what they see as their vision of the future of AI coding agents and how people are gonna use them over the next year and beyond.
01:51And AMP took their previous coding agent. They almost completely scrapped their entire product and rebuilt the entire coding agent from the ground based on first principles and what they saw as the future of AI coding. And if you wanna explore some ways of leveraging this first big shift in vibe coding and being able to go from AI coding inside of an editor to AI agents that are able to code anywhere, there are few things that you can start using that you can start looking at in order to improve the workflow that you already have.
02:21Tools like remote control from ClaudeCode, Cloud Agents from Cursor, and even the Replit and Lovable mobile apps.
02:29And not only that, we're also seeing connections happening between different environments like the terminal and your phone. So you're able to kick off a coding agent inside of the terminal and then carry on working with that coding agent on your phone until you get back to your desk and you can continue in the terminal.
02:45So these sessions for AI coding agents are actually transferable across different environments and different devices as well. This means that you're not just tied to coding at your desk on your laptop, but if you wanna take your coding sessions elsewhere, whenever you go out, whenever you go to lunch, whenever you're working away from your desk, you can do that with these tools.
03:04So wherever you're vibe coding or coding with AI agents, look at the features and the options that they have there to use cloud based agents in order to continue your vibe coding work when you're not sat at your desk inside of an editor. The second big shift that is happening in vibe coding is going from monitoring your own context to context auto compaction.
03:24And if you've spent any time in AI coding tools, you know that that context bar fills up pretty quickly and you have to keep on top of the context that is going into your chat so that the results you're getting from your AI coding agent don't suffer and the output quality isn't super low. But that is soon gonna be a thing of the past.
03:42All of the frontier models and all of the leading AI coding agents in the world right now are developing systems where they can auto compact memory and context in a conversation in a way which maintains the critical parts of your conversation, doesn't use up a load of tokens, and also doesn't suffer from which happens over an extended session where you have context compacted on top of each other multiple times and you lose all of that critical information that you need in the work that you're doing in that session.
04:11Now this is one of the things that AMP's new AI coding agent is claiming to be able to fix with this auto compaction. And we're gonna see more and more of these frontier models and AI coding harnesses adopt auto compaction as a default technique to maintaining context throughout a long conversation without the user having to keep constantly starting new conversations or new threads, which also presents its own problem because that means the agent has to go and reread the the code base in order to get an understanding of what you're actually building and working on every time you start that new conversation.
04:43And so auto compaction is just gonna make it easier and easier to continue a long running thread or series of chats where you don't have to worry about managing the context yourself. The AI agent is gonna be able to do that. So a great way to think about how you can use this today is are you spending a lot of time on context management?
05:01How can you reduce the amount of time you're spending on context management? And how can you start to trust these AI agents with auto compaction a lot more than you used to, and to be aware that when these things come out and when these features are released in these AI coding agents, they're gonna enable you to work even faster without worrying about how you manage the context for those agents.
05:23Big shift number three in vibe coding is going from send, wait, and review to being able to . And what I mean by this is that as AI coding agents get more and more competent and successful at completing these tasks, especially smaller, more straightforward tasks, the easier it is to begin to queue up a lot of prompts, a lot of changes in a row as a sort of backlog or to do list for the AI agent to actually go and complete without worrying too much about context switching or whether the AI agent is actually gonna complete that task at a high enough quality.
05:59Now previously, a lot of best practice for coding and vybe coding with AI coding agents is about making a change, building a feature, and then doing regular code review in order to maintain the quality of the code that is written by the AI agent. And so all that this means in the long run is that you're gonna start to see that window increase, whereas you build with AI coding agents, you're not gonna need to do code review as much.
06:22You're not gonna need to review the code that it's writing as much. And you're not gonna need to keep waiting to actually fix any of the issues that come out of that code. Now, this can combine really, really well with actually a very rigorous bug fixing and code review process.
06:36And you can even automate this where if you're working on a lot of codes during the day, if you're building an app or building anything where you're vibe coding with these coding agents, to set up automations where at the end of each day, it automatically goes and reviews the code that it's written that day, identifies any bugs, plans how to fix them, and then fixes them for you before you start work the next day on the next set of features that you wanna build.
06:59And you can see this playing out inside of the new Codex desktop app from OpenAI, where when you submit a message to the coding agent and then you send follow-up messages on top of that, it actually stacks them in a line on top of each other where you can drag those messages around inside of the queue and you can send those messages straight to the agent directly to interrupt its chain of thought and actually steer the AI agent in the process that it's already working on.
07:24And if you wanna start using this today, then start using something like the Codex desktop app and start getting comfortable with submitting those series of tasks and prompts on top of each other in a queue that the AI agent can work through one after the other without you having to directly supervise every single thing that it's doing.
07:41Big shift number four in vibe coding is going from approving every request and every permission to . And there are people who are already doing this months ago like Peter Steinberger, Riley Brown, and many others in the vibe coding and AI coding space where they were already using dangerously skipped permissions pretty much a 100% of the time.
08:02And the reason that this is happening is because of two things. First is the increase in the quality of the reasoning of the AI coding agent and that it can actually make better decisions as it's going through step by step trying to identify what it needs to do in a plan and actually build out the feature or the application that you're working on for you.
08:20And the second reason this is happening is because now with sub agents and all of these different processes and reasoning steps that coding agents are running, there are just a vastly increased number of tool calls and permission requests to the point where if you are manually approving those permissions, you can just sit there hitting yes yes yes yes approve constantly until the AI agent actually completes its work and 99% of those approvals actually didn't need your input at all.
08:48And this is part of the reason why Anthropic and OpenAI have introduced the auto review modes as part of their coding agents. And the way that you can start using this today is to start working with these auto review modes for these coding agents to actually lower the permission levels that you have set as default in the coding tools that you're using and to trust the agent and the model more in terms of what it's actually doing.
09:11And the final shift here in Vibe coding is is going from one project and one stream to multiple projects and multiple agents running at the same time. And part of being able to do this is actually just a skill issue in terms of how you use these AI coding tools. It's about building up the confidence to give these AI coding agents a bit more free rein in the tasks they're doing, to feel comfortable working across multiple projects, multiple folders at the same time, and to understand how you can orchestrate multiple agents across different sets of tasks where they aren't gonna interfere with each other in terms of writing code to the same files and features inside of your code base or creating conflicts and bugs that contradict each other.
09:52An easy way to start doing this is to think about what tasks can I do inside of my project that aren't necessarily to do with writing code? Maybe I wanna create a launch video for my app idea, and I actually do that inside of my code base because the agent will have context of the app that I'm actually building, and then you can use a skill like hyperframes to be able to build that launch video for you.
10:12You could also do things like conducting research, building context files, and turning code into a number of other assets including websites, social posts, business strategies, presentations, spreadsheets, and a whole load of other things.
10:26And you can do all of this inside of these AI coding tools, especially now with tools like the new Codex desktop app from OpenAI, which is as good of a tool for just general purpose knowledge work as it is for writing any code or building software and application. And so how you can get started with this today is to start using something like the OpenAI Codex PAP and start running multiple agents in multiple projects with different types of tasks.
10:49And not only that, but you can think about how do I turn the projects that I'm working on into a folder on my computer which have a set of markdown files that give the agent context of what it needs to build, and how do I install the right skills into those projects to allow the agent to be able to do a whole set of other tasks like creating video, editing video, creating images, social posts, building websites, and all of the stuff that I mentioned previously.
11:15And so those are the five big shifts that are happening in VIVE coding and AI coding right now. And so if you're building with AI, you need to be aware of these things. What is happening in the AI coding space and how to adjust and change your workflows so that you're building skills are not only gonna be relevant now, but are gonna be relevant for the future of AI coding.
11:34Now I'd love to know which shift you think is gonna be the biggest of these and if you're leaning into any of these shifts already. And if you are building with AI and you wanna build apps and software that actually work and launch to real customers, I've got a community helping people do just that over at school.com/aiapps.
11:49You can find out more by clicking the link in the description down below. There's over 250 people inside that community who are building real applications and have launched to real customers. If you enjoyed the video, don't forget to like and subscribe.
12:01Thank you for watching, and I will see you next time.
The Hook
The bait, then the rug-pull.
Twelve minutes. One camera. No B-roll. And a host who opens by declaring the genre he is talking about dead. Chris from Build Great Products is not being contrarian for sport. He is tracking five structural shifts he thinks most AI builders are sleeping on, and his case lands because he names actual tools, actual people, and one embarrassingly relatable image: sitting there hitting yes yes yes yes approve while the agent just works.
Named ideas worth stealing.
00:15list
The Five Shifts of Vibe Coding
- Agents code anywhere (not just inside an IDE)
- Auto-compaction replaces manual context management
- Queue-stacking replaces send-wait-review
- Permission delegation - let the agent cook
- Multi-agent multi-project orchestration
A sequential framework mapping the behavioral evolution of AI-assisted coding from 2024 babysitting to 2026 fleet orchestration.
Steal forAny state-of-AI content, tutorial on upgrading your Claude Code or Cursor workflow, or positioning piece on why solo builders are now faster than dev teams
How they asked for the click.
VERBAL ASK
11:37product
“If you are building with AI and you wanna build apps and software that actually work and launch to real customers, I have got a community helping people do just that over at school.com/aiapps.”
Soft sell - mentioned twice (60s and 11:49). No hard pitch, no price, just community size (250+) and outcome promise. Low friction.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
