Agentic Development with Local AI
Agentic Development with Local AI
Deploy and run a reliable agentic development workflow using local models on your own hardware. In one day!
Successful agentic development means directing an AI coding agent through a disciplined process.
Doing it on local infrastructure keeps your specs, your codebase, your context, and your data on your machine.
This workshop teaches that workflow, and proves it by having you deploy something real.
WHO SHOULD ATTEND
Anyone concerned about their dependence on cloud AI and ready to do something concrete about it. Be it economic, data security, or data sovereignty. Product managers, founders, consultants, engineering leads, analytics and ops managers, and technical staff in regulated environments all fit this room.
Your reason is yours: cost control, privacy, model stability, data sovereignty, client confidentiality, or a compliance officer asking questions you cannot yet answer. Bring the reason and a problem worth solving.
You must already have experience and comfort with agentic development practices. This is not an introductory course on agentic AI.
What you leave with:
Experience deploying and configuring a local open weights model(s) on your hardware or private cloud instance.
Understanding on how to tune your model and make it part of a reliable development workflow.
How to combine local models with foundational proprietary models to get the best of both worlds and keep your token costs to a minimum.
The ability to address data sovereignty and security issues by keeping your AI on premises.
A repeatable agentic development workflow running on infrastructure you control
Deeper understanding of agentic prompting and methodologies for working with the broader LLM ecosystems.
Understanding of hardware requirements and limitations of local models and how to adapt to them.
A project in active development on your local system.
A defensible framework for the local-versus-cloud decision, with the cost and risk case for both
One year of 021:School, plus membership in our agentic AI professional community.
Why this, why now
Soon, you will need to clearly answer:
Where does your codebase go once it leaves the building?
Why have your SaaS bills tripled, what control do you have on it?
How do you ensure compliance with customer data security and sovereignty requirements?
Which jurisdictions can access your data and context?
What happens when Anthropic, OpenAI, et al gets acquired, IPO, raises prices, or quietly restricts the capability you build upon?
What you build
You define it: If you're an experienced agentic developer - bring your own project. We will vet it for you in advance.
We define it: For those still developing their skills with Agentic AI, we will provide you with a project so you can focus more on model tuning and workflow.
Time Block
09:00: The local stack - what it does today and what it opens up
10:00: Live teardown of a working local AI app, including the bug that crashed it
10:30: Verification - watch the model run on your own silicon
11:00: Scaffold your spec set
13:30: Build. Ninety minutes, hard ceiling
15:00: Deploy and debugging
15:45: Orchestrating local models with either OpenAI or Anthropic models.
16:45: Questions, discussion
Your Instructors
Alexandre Morin:
Alexandre is an AI-product builder and product management leader with a background in mobile device management and security. He builds enterprise tools with user experiences that change how people work, and is known for bringing local AI solutions to organizations that care about privacy and data sovereignty.
Ex-Apple, he was previously Product Management Lead at Mosyle and spent three years at Kandji as a Senior Product Manager. Today he runs Borealis, and works with Apple as an independent contractor on AI product building and local AI reference architecture. He recently relocated from San Francisco to Vancouver.
David Gratton:
David is the founder of 021:Events, co-founder of NorBot.ca and the Agentics Foundation. He is a 25 year software and product veteran covering AI, video games, SaaS, B2B, robotics, and consumer products.
**Hardware. Read this before you register.**
Mac: Apple Silicon, 24GB unified memory minimum, 32GB recommended. Perfect for a modern Mac Mini. NOTE: Intel Macs will not work.
Linux: NVIDIA GPU with 16GB VRAM minimum, 24GB recommended. 32GB system RAM. Ubuntu 22.04 or newer with current NVIDIA drivers. AMD, Intel Arc, and integrated graphics are not supported this session.
Windows is NOT SUPPORTED for this session.
Cloud Access: If you do not have the hardware to bring for the above, we can help you set up on a private cloud instance.
Linux attendees confirm their seat by posting verification test output in Slack. We send the test with the pre-work, and a facilitator helps you if it fails.