Domesticating AI

Hosts: Miriah Peterson, Matt Sharp, Chris Brousseau
Recorded: April 2026
Status: Released

Most AI systems today are designed to be helpful — not secure.

In this episode, we break down how AI systems actually get exploited in production:

• a real supply chain attack on a widely used AI dependency
• prompt injection and why it still works
• image-based (multimodal) exploits
• tool and agent abuse

If you’re building AI — especially at a startup — you are the security team.

A widely used AI dependency was compromised via a malicious .pth file:

• executes automatically when Python starts
• no import required
• targets credentials, SSH keys, and environment variables

👉 Just installing the package was enough.

This highlights a critical reality:

Your AI system is only as secure as your dependencies.

• Models cannot distinguish between instructions and data
• External content can override system behavior
• Still one of the most common AI vulnerabilities

🔗 https://learnprompting.org/docs/prompt_hacking/injection

• Hidden instructions embedded in images
• AI interprets images differently than humans
• Expands the attack surface significantly

🔗 https://arxiv.org/abs/2306.11698

• AI systems can take real-world actions via tools
• Prompt injection → API calls, data leaks, unintended execution
• Agents amplify risk through autonomy and retries

If you’re building AI systems today:

• separate instructions from data
• limit tool permissions
• treat outputs as untrusted
• validate everything before execution

• AI systems have an internet-sized attack surface
• Supply chain attacks bypass all AI safeguards
• Prompt injection is a fundamental problem
• AI doesn’t fail safely — it fails wherever your system is weakest

• LiteLLM incident: https://github.com/BerriAI/litellm/issues/24512
• Attack breakdown: https://futuresearch.ai/blog/litellm-pypi-supply-chain-attack/
• LLM attack techniques: https://llm-attacks.org/
• OWASP LLM Top 10: https://owasp.org/www-project-top-10-for-large-language-model-applications/
• Gandalf challenge: https://gandalf.lakera.ai/

We’ve launched a Patreon for Domesticating AI 🎉

Get:

• early access to episodes
• behind-the-scenes content
• bloopers and uncut moments

👉 https://patreon.com/DomesticatingAIPodcast

• 🎥 YouTube: https://youtu.be/HTTxE7Y1sko

What’s the weirdest way an AI system has broken for you?

Keep your AI on a leash.

AI can write code — but that doesn’t mean you should trust it.

In this episode of Domesticating AI, we’re joined by Tyler Folkman (author of The AI Architect) to break down how engineers are actually using AI to build software — and why most people are still just vibe coding.

• Vibe coding vs real engineering
• Reasoning models vs coding models
• How to plan and prompt AI effectively
• When to let AI take the wheel (and when not to)
• Local vs cloud coding agents
• Token costs vs owning hardware

• Tyler Folkman — The AI Architect
  

• Anthropic
  https://www.anthropic.com
• OpenAI
  https://openai.com
• Ollama
  https://ollama.com
• MiniMax-M2.5
  https://ollama.com/library/minimax-m2.5
• GLM-5
  https://ollama.com/library/glm-5

• AmpCode Chronicle
  https://ampcode.com/chronicle
• Andrej Karpathy on Context Engineering
  https://x.com/karpathy
• “Human in the Loop is Tired”
  (add link if you have it)

Domesticating AI is a bi-weekly podcast about practical AI for developers.

We help you brace the feral open-source AI landscape — so you can tame it instead of getting dragged by it.

[email protected]

Spotify
https://open.spotify.com/show/2WsAR4fvcXzp3vVZGVlkE2

Apple Podcasts
https://podcasts.apple.com/us/podcast/domesticating-ai/id1873338950

Are you vibe coding — or engineering with AI?

Let us know your setup.

Keep your AI on a leash.

🧠 What We Cover🔗 Links & ResourcesGuestModels & ToolsArticles / Mentions🎧 About the Podcast📬 Contact🔥 Follow👇 Join the Discussion

Recording Date: February 27, 2026
Hosts: Miriah Peterson, Matt Sharp, Chris Brousseau

Running AI locally is easier than ever.
Running it securely is another story.

In this episode of Domesticating AI, we break down the moment every homelab builder hits:

The second you move from one machine to two machines…
access becomes your first real engineering problem.

We explore the real architecture questions behind self-hosting AI:

• Why a dedicated machine isn’t a sandbox

• Why Docker alone isn’t isolation

• How homelabs evolve from Plex servers to AI infrastructure

• The blast radius problem with local agents

• Why networking and access control matter more than model size

We also discuss the surge in local AI hardware demand and the risks of running powerful agents on machines with unrestricted access.

Whether you're running OpenClaw, Ollama, a NAS, Postgres, or a home automation stack, the same rule applies:

Infrastructure without containment is just risk waiting to happen.

High-memory Mac Minis are seeing long shipping delays as developers rush to build local AI systems.

https://www.tomshardware.com/tech-industry/artificial-intelligence/openclaw-fueled-ordering-frenzy-creates-apple-mac-shortage-delivery-for-high-unified-memory-units-now-ranges-from-6-days-to-6-weeks

Marketplace plugins and execution boundaries are becoming a growing security concern in agent systems.

https://www.linkedin.com/posts/matthewsharp_i-use-to-do-nothing-but-post-about-clean-activity-7432832983339999232-iR04

Overview of risks around agent plugin ecosystems and execution boundaries.

https://conscia.com/blog/the-openclaw-security-crisis/

Private mesh networking used to securely access homelabs.

https://tailscale.com

Local AI coding agent framework.

https://openclaw.ai

Local LLM runtime used for running models on personal machines.

https://ollama.com

• Why people actually build homelabs

• Plex, NAS, and home automation as infrastructure entry points

• AI workloads vs dev workloads

• Why long-running services shouldn’t live on your laptop

• Networking architecture for homelabs

• RBAC-style access control between machines

• Secrets management mistakes developers make

• Containment and blast-radius thinking for AI agents

• Tailscale and private mesh networking

Each host answers:

If I had $0

• What I would run

• What I would avoid

If I had $1K

• What machine I’d buy

• How I’d isolate workloads

If I had $5K

• How I’d segment infrastructure

• What monitoring I’d deploy

• What I would never expose to the internet

Staff Data Engineer, content creator, and founder of SoyPete Tech.
Miriah focuses on practical AI systems, Go infrastructure, and self-hosted AI engineering.

She is also a Google Developer Expert in Go and organizer of Go West Conf.

https://soypete.tech

AI engineer and co-author of LLMs in Production.
Matt focuses on applied AI systems, local model infrastructure, and developer-focused AI tooling.

Software engineer and AI practitioner focused on practical applications of machine learning and developer infrastructure.

Domesticating AI is supported by the SoyPete Tech community.

If you enjoy the show:

• Subscribe on YouTube

• Follow on Spotify

• Join the Discord community

• Share the episode with another engineer building with AI

More content and tutorials:

https://soypetech.substack.com

📰 News DiscussedMac Mini Shortages from Local AI DemandOpenClaw Security DiscussionOpenClaw Security Concerns (Referenced)🧰 Tools & Technologies MentionedTailscaleOpenClawOllama🏗 Topics Covered⚡ Lightning Round🎙 HostsMiriah PetersonMatt SharpChris Brousseau🤝 Sponsors

📅 Recorded: February 6, 2026

In this episode of Domesticating AI, we discuss why scaling AI systems with more compute often hides weak engineering decisions — especially in agent workflows. We explore constrained hardware, context management, tool calling, logit manipulation, and why small models can make you a better AI engineer.

• Moltbot / Clawdbot overview (The Verge)
  https://www.theverge.com/report/869004/moltbot-clawdbot-local-ai-agent

• Fake Moltbot VS Code extension spreading malware
  https://thehackernews.com/2026/01/fake-moltbot-ai-coding-assistant-on-vs.html

• Exposed Moltbot admin panels and credential leaks
  https://www.bitdefender.com/en-us/blog/hotforsecurity/moltbot-security-alert-exposed-clawdbot-control-panels-risk-credential-leaks-and-account-takeovers

• Cloudflare Moltworker (self-hosted agent on Workers)
  https://blog.cloudflare.com/moltworker-self-hosted-ai-agent/

• LangGraph – https://www.langchain.com/langgraph

• LangChain – https://www.langchain.com/

• Langfuse – https://langfuse.com/

• Pydantic AI – https://github.com/pydantic/pydantic-ai

• Instructor – https://github.com/jxnl/instructor

• Hugging Face SmolAgents – https://huggingface.co/blog/smolagents

Have a topic suggestion or want to sponsor the show?

📩 [email protected]

Episode 3 is a hardware-first guide to running AI at home. We break down what CPUs vs GPUs vs NPUs vs TPUs actually do in the inference pipeline, why memory capacity isn’t the same as performance (model loading, KV cache, and MoE), why backends/runtimes are real constraints (CUDA vs ROCm vs Metal/MLX vs CPU), and how to scale from one box to multi-GPU and multi-machine setups.

Keep your AI on a leash.

Links mentioned:

- GPU Glossary (Modal): https://modal.com/gpu-glossary

- CUDA → ROCm headline: https://wccftech.com/the-claude-code-has-managed-to-port-nvidia-cuda-backend-to-rocm-in-just-30-minutes/

- Unsloth PR: https://github.com/unslothai/unsloth/pull/3856

Everyone keeps calling NVIDIA DGX Spark an “inference box”… but in practice it behaves more like a dev rig.

In Ep 2 of Domesticating AI, we break down what Spark is actually good for (AI development + fine-tuning) vs what it isn’t (a magical drop-in inference server). We also dig into why unified memory changes the local-AI experience, the “gateway stack” (Ollama + Open WebUI), when you outgrow turnkey UIs, and how homelab economics + networking decisions shape what you should run at home.

In this episode

• Training vs inference (and why “inference server” gets misused)

• Unified memory: what it changes for model loading + workflows

• Ollama + Open WebUI as the fastest on-ramp for local AI

• Fine-tuning workflows (QLoRA/Unsloth-style) and where Spark shines

• Homelab reality: Docker “recipes,” troubleshooting, and collaboration

• Safer remote access: Tailscale

• Cloud vs home economics (when cloud is cheaper… and when it explodes)

NVIDIA / DGX Spark

• DGX Spark: https://www.nvidia.com/en-us/products/workstations/dgx-spark/

• Build hub / recipes: https://build.nvidia.com/spark

• NIM on Spark playbook: https://build.nvidia.com/spark/nim-llm

Local AI runners + UIs

• Ollama: https://ollama.com/

• Open WebUI (GitHub): https://github.com/open-webui/open-webui

• Open WebUI docs: https://docs.openwebui.com/

• llama.cpp: https://github.com/ggml-org/llama.cpp

• LM Studio: https://lmstudio.ai/

• vLLM: https://github.com/vllm-project/vllm

• Jan: https://jan.ai/

Fine-tuning + workflows

• Unsloth: https://github.com/unslothai/unsloth

Image generation tools (mentioned)

• ComfyUI: https://github.com/Comfy-Org/ComfyUI

• AUTOMATIC1111 SD WebUI: https://github.com/AUTOMATIC1111/stable-diffusion-webui

Networking / Remote access

• Tailscale: https://tailscale.com/

Cloud GPU alternatives (mentioned)

• Runpod pricing: https://www.runpod.io/pricing

• Modal pricing: https://modal.com/pricing

Miriah Peterson (Host): Miriah Peterson is a software engineer, Go educator, and community builder focused on production-first AI—treating LLM systems like real software with real users. She runs SoyPete Tech (streams + writing + open-source projects) and stays active in the Utah dev community through meetups and events, with a practical focus on shipping local and cloud AI systems.
Connect:

• SoyPete Tech (YouTube): https://www.youtube.com/@SoyPete_Tech

• SoyPete Tech (Substack): https://soypetetech.substack.com/

• LinkedIn: https://www.linkedin.com/in/miriah-peterson-35649b5b/

Matt Sharp (Host): Matt Sharp is an AI Engineer and Strategist for a tech consulting firm and co-author of LLMs in Production. He’s a recovering data scientist and MLOps expert with 10+ years of experience operationalizing ML systems in production. Matt also teaches a graduate-level MLOps-in-production course at Utah State University as an adjunct professor. You can find him on Substack (Data Pioneer), LinkedIn, and on his other podcast, the Learning Curve.
Connect:

• Data Pioneer (Substack): https://thedatapioneer.substack.com/

Chris Brousseau (Host): Chris Brousseau is a linguist by training and an NLP practitioner by trade, with a career spanning linguistically informed NLP, modern LLM systems, and MLOps practices. He’s co-author of LLMs in Production and is currently VP of AI at VEOX. You can find him as IMJONEZZ (two Z’s) on YouTube, GitHub, and on LinkedIn.
Connect:

• YouTube (IMJONEZZ): https://www.youtube.com/channel/UCPtkaw_x97yP4WevW7axk0g

• LinkedIn: https://www.linkedin.com/in/chris-brousseau/en

📘 LLMs in Production (Matt Sharp & Chris Brousseau): https://www.manning.com/books/llms-in-production

Links & ResourcesHosts

Domesticating AI — S01E01: Your First AI at Home

Hosts: Miriah Peterson, Matt Sharp, Chris Brousseau

This episode is your practical on-ramp to running AI at home: why inference engines matter, what to install first, and how to make “local AI” feel stable instead of fragile. The hosts start with a hardware + market reality check (tinygrad’s tinybox-style “AI server appliance” idea and the ongoing memory/RAM crunch), then break down what an inference engine actually does, how popular runtimes compare (llama.cpp, vLLM, Ollama, TGI), and a sane starter workflow for getting from “downloaded a model” to “usable local AI.”

• ​Inference engines are the “runtime”: model loading, tokenization, KV cache/context handling, and the serving layer.
• ​Pick your engine based on your goal: tinkering (llama.cpp) vs serving throughput (vLLM/TGI) vs it-just-works packaging (Ollama).
• ​You don’t need a brand-new rig to start, but RAM/VRAM constraints will shape everything.
• ​Use leaderboards as a hint, then validate with your own small eval prompts that match your workload.
• ​If you’re exposing anything beyond your LAN: reverse proxy + TLS + don’t casually open ports.

0:00 Intro + host chaos + what the show is

1:08 News: tinygrad / “AI server appliance” thinking (tinybox vibes)

2:44 News: RAM prices + the memory crunch for builders

8:26 Main: building your first AI at home (why now)

8:49 What is an inference engine?

12:30 Engines compared: llama.cpp vs vLLM vs Ollama vs TGI

15:42 Do you need to buy a new computer? (CPU vs GPU realities)

25:32 Models for home: fit-to-hardware, quantization, context

34:37 Leaderboards vs evals: picking models you can trust

44:00 Community + meetups + where to follow

45:22 Outro — “Keep your AI on a leash”

News / context

• ​Tom’s Hardware: TinyBox production + multi-GPU appliance concept (Tom's Hardware)
• ​Reuters: AI-driven memory shortage / supply-chain crunch (Reuters)
• ​IDC: 2026 device impacts from the memory shortage (IDC)

Inference engines

• ​llama.cpp (GGML org) (GitHub)
• ​vLLM OpenAI-compatible server (docs.vllm.ai)
• ​Ollama docs (quickstart) (Ollama Documentation)
• ​Hugging Face Text Generation Inference (TGI) (GitHub)
• ​Miriah Peterson: Software engineer, Go educator, and community builder focused on production-first AI. Runs SoyPete Tech (streams + writing + open-source).
• ​Matt Sharp: AI Engineer/Strategist, co-author of LLMs in Production, MLOps practitioner. Writes The Data Pioneer. (thedatapioneer.substack.com)
• ​Chris Brousseau: NLP practitioner, co-author of LLMs in Production, VP of AI at VEOX. You can find him as IMJONEZZ. (veox.ai)
• ​SoyPete Tech (YouTube): (youtube.com)
• ​SoyPete Tech (Substack): (soypetetech.substack.com)
• ​Matt’s Substack (The Data Pioneer): (thedatapioneer.substack.com)
• ​Chris on YouTube (IMJONEZZ): (youtube.com)
• ​LLMs in Production (book): (Manning Publications)