Method · MoneyNnenna Ndukwe maintains planning templates that apply to any repo, defining the expected plan structure regardless of the codebase. The planning template includes a summary, problem statement, constraints, scope, behavior scenarios, verification gates, failure and recovery rules, a commit plan, and skill routing to pre-plan which skills the coding agent will kick off during implementation. The build execution template documents the starting state (what branch, current local verification process, expected commands) and includes a definition of done or completion note. Document everything in the plan including deferred items and changed decisions.
Endorsed; no stated uncertainty.
Method, not a verified protocol.
AI Engineer · How to Build Quality Gates into Agentic Coding Workflows — Nnenna Ndukwe, Qodo AI
Method · MoneyRun threat modeling scans on every code or AI application change pushed (e.g., pull requests). Set up webhooks that trigger a scan whenever something changes, run a scan whenever infrastructure-as-code is changed, and monitor Terraform files created by agents — running a scan the moment they change. Integrate threat modeling early when designing AI-agentic workflows.
Method, not a verified protocol.
AI Engineer · AI Security Engineer Foundations + Certificate — Javier Garza, Snyk
Recommendation · MoneyThe speaker recommends using automated tools to check that any MCP server you install is free of malware before running it for the first time. He admits he does not personally check all the code of every MCP server he installs, but he limits a lot what he installs.
AI Engineer · AI Security Engineer Foundations + Certificate — Javier Garza, Snyk
Recommendation · MoneyProvide guidance to the agent at the start of optimization to steer it from the beginning. Point the WeCo agent to a markdown file containing all your information and context so the agent has the best chance at finding success. For MiniPG optimization, this can include hard constraints and promising ideas like focusing on compression, different quantization methods, and weight sharing across model blocks.
AI Engineer · Parameter Golf with AutoResearch — Vayum Arora, Zhengyao Jiang, Dixing Xu & Dhruv Srikanth, Weco AI
Recommendation · MoneyShubhankar Srivastava recommends: if you want to ship skills in production, bake them into your agent loop. Build systems where as soon as you run into any production failure, you run an auto research loop to see what went wrong, heal it, and create a pull request that goes directly to a human engineer who fixes it, and the skill gets updated automatically. Alternatively, run a sample run every two or three hours to see if the skill is working perfectly — there are different ways of having cron jobs check skill accuracy. On Anthropic managed agents, run a cron job using the skill every day at 9 a.m., configured to send a notification on failure, which triggers a healing loop to handle websites changing or anti-bots changing — baking in checks and balances around changing websites and antiB
AI Engineer · Hill-Climbing Skills: Improve Agents Without Changing the Model — Shubhankar Srivastava, Browserbase
Recommendation · MoneyA proposed approach: replicate applications (like world models) so you have better control and guardrails, then let the agent do free exploration in the environment about 100 times so it becomes familiar with the environment. After exploration, the goal is finding the first successful trajectory; once found, it can be successful all the time. For an initial run on a program the agent has never seen (e.g., a legacy payroll process), the initial prompt has to be very specific, e.g., go to an employee, go to payroll, check this box.
proposed approach; 'a couple of months' out for the current product
AI Engineer · The Autonomous Computer: Infrastructure for Computer Use Agents — Ang Li, Simular
Recommendation · SocialCoach Kyle recommends starting conversations with strangers by asking a simple, low-pressure question about something in the environment, calling it the 'obvious question' technique. He says it works in any setting and is one of the easiest first steps toward becoming more social.
Coach Kyle · How to make SMALL TALK so natural, strangers start talking to you
Recommendation · SocialCoach Kyle recommends committing to giving out compliments and greetings to at least three people per day. He frames comfortable greetings and compliments as the foundational level-one step of his social roadmap, and says taking this one step consistently can open many doors and build social connections.
Coach Kyle · How to make SMALL TALK so natural, strangers start talking to you
Recommendation · MoneyDhruv Nathawani recommends giving the agent a single tool (Tavily Search) in the system prompt, adding usage guidelines so the agent does not over-call the tool and rack up large costs, and including an in-prompt example so the model learns how to call the search tool via in-context learning.
AI Engineer · Teaching Agents to Search with NVIDIA Data Designer — Dhruv Nathawani, NVIDIA
RecommendationOmar urges the audience to send the Epstein cover sheet to ChatGPT and try to get results out of it, to see how accurate (inaccurate) frontier models are on complex layouts.
AI Engineer · Parsing the Epstein Files for JMail — Palak Agarwal & Omar Alhait, Reducto
Recommendation · MoneyThe agent hallucinates when a tool fails instead of disclosing the failure, so prompt-based changes are needed in addition to evals for tool failure. Duggal recommends building an eval on top of it to detect when the tool is failing AND making prompt-based changes so the agent doesn't hallucinate. He also suggests instructing the agent to build the tool-success/silent-failure eval and fix numeric groundedness so responses are always grounded to an actual reference point instead of made-up numbers.
AI Engineer · Let Your Agent Cook: Using Skills to Evaluate and Improve Your App — Ankur Duggal, Arize AI
Method · MoneyNyah Macklin shows how to set the fixed size splitter to chunk size 500 with chunk overlap, define sample node types (technology, concept, example, process, challenge, benefit), define relationship types (related to, part of, used in), and structure the graph with node-relationship-node patterns.
Method, not a verified protocol.
AI Engineer · RAG Needs a Map: Using GraphRAG to Retrieve Connected Context — Nyah Macklin, Neo4j
Method · MoneyNyah Macklin outlines the overall knowledge graph construction process: gather data sources, chunk the data, vectorize it, extract nodes and relationships using an LLM (prompting for JSON output), parse the JSON, and generate the graph — noting that LLM extraction cuts down time versus doing it manually.
Method, not a verified protocol.
AI Engineer · RAG Needs a Map: Using GraphRAG to Retrieve Connected Context — Nyah Macklin, Neo4j
Recommendation · BodyDo 10 sets of 10 calories or one set of 100 calories on the air bike every week or twice a week, done pretty intensely, and you will improve your conditioning without a doubt. You can get an excellent quality conditioning session in less than 10 minutes that makes genuine progress.
Garage Strength · How To Get Strong and Still Be Athletic w/ Sika Strength
Recommendation · MoneyThe speaker recommends creating an agent that looks inside the code base, identifies opportunities for deduplication of code, and creates a series of issues for you.
AI Engineer · From Your Laptop to the Pipeline: Scaling Custom Agents with GitHub Copilot
Recommendation · MoneyThe speaker recommends using Orchestrate when modifying an API in a microservice that requires a downstream change, to make cross-repo changes simultaneously with several agents.
AI Engineer · From Your Laptop to the Pipeline: Scaling Custom Agents with GitHub Copilot
Recommendation · MoneyGabriel Martinez recommends breaking features into small atomic pieces small enough to hold in your head, rather than requesting a whole feature set in one shot without understanding what you're asking for. Decomposing needs into the simplest chunks lends to understanding of simpler code surfaces and control of code bases.
Creator advice attributed to Gabriel Martinez.
AI Engineer · Generation Is Cheap, Review Is Expensive: How to Stop Shipping AI Slop — Gabriel Martinez, G2i
Recommendation · MoneyGabriel Martinez recommends centering rewards around producing value, elegant well-factored pragmatic system design, and clarity of tests and evaluation harnesses, and making accountability metrics visible for each chunk of work product.
Creator advice attributed to Gabriel Martinez.
AI Engineer · Generation Is Cheap, Review Is Expensive: How to Stop Shipping AI Slop — Gabriel Martinez, G2i
Recommendation · MoneyGelsinger endorses moving datacenters to 800-volt DC power delivery to eliminate the many historical AC-to-DC conversion steps in the power network — 'the revenge of Edison'. He also calls for rebuilding the entire power delivery network with solid-state transformers and switches to unlock innovation in power infrastructure, and for the US to get much more efficient at delivering the power network including expanding domestic supply chains.
The a16z Show · Chips, Memory, and Power | Pat Gelsinger
Recommendation · MoneyCisco guaranteed its employees that if they don't use AI, they will lose their job — the opposite of threatening that AI use would cost jobs. Patel says if you're not using AI, chances are Cisco is the wrong place for you, so go all in and the company will give you all the tooling and support. He notes that if someone did not use a PC or the internet, you would never hire them in a knowledge worker job, so not using AI in a secular shift is similarly disqualifying.
Y Combinator · How Jeetu Patel Runs Cisco Like the World’s Largest Startup