Recommendation · MoneyJavier Garza recommends taking a prompt engineering class or course if you haven't already — there are many free ones; you can search YouTube for 'prompt engineering tutorial.' He suggests spending two days doing three to five prompt engineering classes to learn how to interact with agents and avoid repeated tuning. When crafting a prompt for an agentic tool, tell the tool explicitly what you want using prompt engineering techniques. He shares that he did five prompt engineering courses over Christmas during a week off and considers it one of the best investments he made last year.
AI Engineer · AI Security Engineer Foundations + Certificate — Javier Garza, Snyk
Recommendation · MoneyJavier Garza recommends that before publishing an app you plan to sell or put in an external repo, you at least run a security scan and have a developer friend review it for big issues, so you don't get into trouble afterwards.
AI Engineer · AI Security Engineer Foundations + Certificate — Javier Garza, Snyk
Recommendation · MoneyJavier Garza recommends running scans and threat modeling checks on every change pushed — pull requests, code changes, infrastructure-as-code changes, data pipeline updates, and AI model deployments — triggered via webhooks. If an agent creates a Terraform file for cloud infrastructure, monitor that file so a scan runs the moment it changes.
AI Engineer · AI Security Engineer Foundations + Certificate — Javier Garza, Snyk
Recommendation · MoneyThe first step in threat modeling is to scope the system: identify architecture, assets, data flow, and trust boundaries.
AI Engineer · AI Security Engineer Foundations + Certificate — Javier Garza, Snyk
Recommendation · MoneyJavier Garza explains that LLMs are like a database without real-time data: asking Gemini for the current weather yields a system error. To get real-time information from a real-time source, you need to use the MCP protocol with a tool that provides the functionality you need. He also notes an agent cannot scan your code for security issues well without an MCP server, and that when a tool lacks a capability you can create an MCP server by asking Claude Code or Copilot: 'Create an MCP server that does this and that' — the LLM generates the code and a JSON config file used to install it into the agent.
AI Engineer · AI Security Engineer Foundations + Certificate — Javier Garza, Snyk
Recommendation · MoneyJavier Garza recommends putting strict firewall rules in place so the AI cannot open ports that should only be internal.
AI Engineer · AI Security Engineer Foundations + Certificate — Javier Garza, Snyk
Recommendation · MoneyThe speaker advises that when crafting a prompt for an agentic tool it is very important to apply prompt engineering; if you have not taken a prompt engineering class or course, just do it — there are a bunch available for free (search YouTube for prompt engineering tutorials). He recommends doing at least two or three courses, spending two days doing three to five classes so you learn how to interact with agents and type things in the right way; learning prompt engineering saves you from spending time tuning prompts over and over. He shares that he did about five prompt engineering courses last Christmas during a week off and considers it one of the best investments he made last year.
AI Engineer · AI Security Engineer Foundations + Certificate — Javier Garza, Snyk
Recommendation · MoneyThe speaker advises the audience to develop security-focused prompt templates that prioritize security coding practices as part of a secure-by-design playbook.
AI Engineer · AI Security Engineer Foundations + Certificate — Javier Garza, Snyk
Recommendation · MoneyJavier Garza recommends using boundaries by separating environments for dev, staging, and production.
AI Engineer · AI Security Engineer Foundations + Certificate — Javier Garza, Snyk
Recommendation · MoneyJavier Garza recommends minimizing the command allow list in AI-native IDEs and CLIs to only what you need, restricting the tools to specific project directories only, and monitoring for unexpected IDE behavior.
AI Engineer · AI Security Engineer Foundations + Certificate — Javier Garza, Snyk
Method · MoneyA locally coherent method for interacting with a live run: hover over a node in the dashboard to get a summary of what changed, then open it and read the exact diff so when the metric jumps you can read the change that caused it; inject a new idea in plain English during the loop, which opens a new branch and develops new nodes and a new curve; and put a budget on each node so the system spends compute where you want.
Method, not a verified protocol.
AI Engineer · Parameter Golf with AutoResearch — Vayum Arora, Zhengyao Jiang, Dixing Xu & Dhruv Srikanth, Weco AI
Method · MoneyFor coding agents, start from a Docker image or a specific EC2 instance so you can replay from different parts of the stack.
Method, not a verified protocol.
AI Engineer · Hill-Climbing Skills: Improve Agents Without Changing the Model — Shubhankar Srivastava, Browserbase
Recommendation · MoneyAfter collecting data from a site, you need to validate data authenticity — e.g., am I getting price back 90 percent of the time, and do I want to validate that I'm getting it back 90 percent of the time — setting up validation rules so data going through your pipeline is checked accurately.
stated
AI Engineer · https://www.youtube.com/watch?v=IK8cC4FScps
Recommendation · MoneyCheck a site's robots.txt before scraping because many sites block ClaudeBot and GPTBot from accessing their data.
stated
AI Engineer · https://www.youtube.com/watch?v=IK8cC4FScps
Recommendation · MoneyResponses to frequently asked queries like 'what's the weather today in New York City' can be cached because you only need one person to ask and you know the weather for the day, avoiding burning tokens or money to go out on the web.
stated
AI Engineer · https://www.youtube.com/watch?v=IK8cC4FScps
RecommendationWhen evaluating agent trajectories, ask: was this the right tool to pick? Were the arguments valid? Was the response grounded in the right context handed off by another agent?
endorsed
AI Engineer · Building Self-Learning Loops for Your Agent — Fuad Ali, Arize AI
Recommendation · MoneyDiscover which data models belong to which business domains by examining the structure of the data — table-level lineage, column-level lineage, shared keys, and shared naming — rather than relying on folder names and tribal knowledge, which is fragile; when the agent can query domain clusters automatically, asking which models are in the revenue domain becomes a query the agent can run rather than a question requiring back-and-forth with a human.
AI Engineer · The Data Context Layer for Agents — Yoni Michael & Brandon Callender, Typedef
Method · MoneyRung zero is the syntactic level: first parse the code you're trying to understand. Use a language-agnostic parser like TreeSitter as the easiest way to parse code; it gives a basic high-level AST breakdown of the parts of a codebase and how things interact.
Method, not a verified protocol.
AI Engineer · The Data Context Layer for Agents — Yoni Michael & Brandon Callender, Typedef
Recommendation · MoneyYou can give your agent an identity with SPIFFE or SPIRE, and once the agent has an identity, you can make policy around what it is and is not allowed to do.
AI Engineer · Build a Platform and Watch It Burn — Michael Forrester, Accenture & Whitney Lee, Datadog
Recommendation · MoneyForrester recommends doing regex filtering for common known bad patterns in the application wrapper before sending anything to an 8-billion-parameter classifier model, arguing there's no reason not to. He adds that putting an application harness in front that also does regex-level filtering catches a lot of threats as well.
None stated.
AI Engineer · Build a Platform and Watch It Burn — Michael Forrester, Accenture & Whitney Lee, Datadog