How I'm Using AI Agents in 2026
Running one or two AI agents locally on your own machine is now table stakes for developers. But what happens when you need five, ten, or twenty agents running simultaneously — on every pull request, every Slack message, every Linear ticket? This video tackles a problem most creators barely acknowledge: the infrastructure gap between «AI agent demos» and production-grade, cloud-based agent orchestration. The setup is intricate, the payoff is automation at scale, and the question is whether the complexity is worth the leap from local tinkering to distributed agent workflows.
Points clés
Running agents in the cloud allows you to spin up multiple agents in parallel, close your laptop, and let them execute tasks asynchronously — a fundamental shift from local, synchronous agent usage.
The «agent start MD» file and project spec serve as critical context anchors, enabling agents to understand the codebase and follow consistent rules without repeated human intervention.
Skills are reusable, repeatable tasks that can be scheduled, triggered via integrations (GitHub Actions, Slack), or invoked on-demand, turning agents into infrastructure rather than interactive tools.
Environment configuration — including Docker images, GitHub repo access, and team-level permissions — is the foundational setup that unlocks cloud-based agent execution.
GitHub Actions integration allows automatic code review or task execution on every pull request, effectively embedding agents into your CI/CD pipeline.
En bref
If you're serious about scaling AI agent workflows beyond local experimentation, cloud-based orchestration with environment isolation, skill reusability, and trigger-based execution is no longer optional — it's the new baseline for complex development tasks.
What Cloud-Based Agent Orchestration Looks Like
A first look at the AWS agent platform and its UI.
The video opens with a tour of the AWS agent platform dashboard, accessible both inside Warp and via a standalone web UI at aws.warp/dev. Users can view all agent runs, inspect state, view the environment (including the Docker container), and connect directly to a running agent session via SSH. This means you can «steer» an agent even though it's not running on your own computer.
The platform enables scheduling agents to run at intervals, creating reusable «skills» (repeatable tasks like front-end validation or back-end testing), and defining environments that specify Docker images, GitHub repo access, and tooling. Integrations with GitHub, Slack, and Linear allow agents to be triggered by external events — for example, automatically reviewing every pull request. The core promise is continuous, distributed agent execution rather than local, on-demand interaction.
Setting Up the Project Foundation
Generate a spec, scaffold directories, and create a GitHub repo.
Create a spec file Paste a detailed prompt describing the entire application (tech stack, features, architecture) and have the agent generate a markdown spec file. This serves as the project's source of truth.
Scaffold the project structure Instruct the agent to create front-end and back-end directories and basic configuration, but defer detailed implementation to later agents.
Create and commit to a GitHub repo Have the agent create a new remote GitHub repository and commit all changes so far. This repo will be the shared workspace for cloud-based agents.
Run slash-init to index the codebase Generate an «agent start MD» file that explains the project and any rules agents must follow. This context file is critical for coherent agent behavior.
Core Workflow: Agents, Skills, and Environments
Running Multiple Agents in Parallel
Spin up three agents simultaneously for front-end, back-end, and testing.
The speaker creates three markdown plan files (one for front-end, one for back-end, one for testing) and then converts each into a skill stored in .agents/skills. Rather than typing out CLI commands manually, they prompt Warp in natural language: «Run three agents, one for the front-end, one for the back-end, and one for testing. Run these in the cloud using the skills that were just generated, and put them in the environment that we created for this GitHub repository.»
Warp automatically spawns three cloud-based agents. Each agent executes its assigned skill in parallel, operates inside the shared environment, and submits a pull request to the main GitHub repo when complete. The speaker can view real-time status in the AWS UI, open a terminal session to any running agent, and review submitted pull requests. This workflow demonstrates the core value proposition: distributing complex tasks across multiple agents without manual orchestration or local resource consumption.
GitHub Actions Integration for Automatic Code Review
Set up a GitHub Action to trigger agent-based PR review.
Copy the GitHub Action template Navigate to the AWS integrations page, select GitHub Actions, and copy the pre-built YAML configuration for automatic code review on pull requests.
Add the GitHub Action to the repo Paste the configuration into a new .github/workflows directory in your project and commit it. Warp can handle this via a natural language prompt.
Add the Warp API key to GitHub secrets Go to your GitHub repo settings, navigate to Secrets and Variables > Actions, and create a new secret named WARP_API_KEY. Generate the key from Warp's platform settings.
Test with a sample pull request Create a test PR. The GitHub Action should trigger an agent to review the code and post feedback. If the action fails initially, re-run after confirming the API key is present.
When to Use Cloud Agents vs. Local Agents
Balance between local control and cloud-based automation.
When to Use Cloud Agents vs. Local Agents
Running agents locally (via «aws agent run» in a new terminal tab) is still useful for interactive, high-touch tasks. But cloud-based agents unlock asynchronous, distributed workflows: you can start an agent, close your laptop, and review results later. This is especially valuable for backend-heavy projects, long-running builds, or tasks triggered by external events (e.g., every Slack message, every Linear ticket). The speaker emphasizes that Warp's design philosophy — centering the UI on the terminal rather than the code editor — reflects the shift toward agent orchestration as the primary developer interface.
Personnes
Glossaire
Avertissement : Ceci est un résumé généré par IA d'une vidéo YouTube à des fins éducatives et de référence. Il ne constitue pas un conseil en investissement, financier ou juridique. Vérifiez toujours les informations auprès des sources originales avant de prendre des décisions. TubeReads n'est pas affilié au créateur de contenu.