Method · MoneyA locally coherent setup sequence for the hands-on workshop: (1) install prerequisites — git, a GitHub account, GitHub CLI, Python, UV, Node.js, NPM, and a coding agent (preferably Codex or Claude Code); (2) clone the repo and run make setup; (3) sign in to app.kodo.ai using your GitHub email and set up the Qodo account so the skills and rules already in the repo are ingested and apply as context to code review; (4) copy-paste the agent prompts from the README to get started with the task. The slides and repo are available at github.com/nnennandukwe.
Method, not a verified protocol.
AI Engineer · How to Build Quality Gates into Agentic Coding Workflows — Nnenna Ndukwe, Qodo AI
Method · MoneyDuring implementation, Nnenna Ndukwe makes sure proper tests are written using TDD and BDD skills, runs local gates including linters and static analysis after implementing a code change, and uses conventional commits and organized PRs so everything is structured the same way as pre-AI team workflows. Codex ran the verification automatically on its own because of the way the repo was set up, but running it manually is still a best practice.
Endorsed; no stated uncertainty.
Method, not a verified protocol.
AI Engineer · How to Build Quality Gates into Agentic Coding Workflows — Nnenna Ndukwe, Qodo AI
Recommendation · MoneyJavier Garza demonstrates filtering AI BOM results by asset type (e.g., agents, of which he had four), checking the issues tab to see vulnerabilities (e.g., a Llama 3 8B model with one critical and two high vulnerabilities) and trace them to their GitHub repos, using the chat feature to filter results across all monitored reports (e.g., 'show me the top five LLMs'), and creating policies for issues found in AI BOM scans.
AI Engineer · AI Security Engineer Foundations + Certificate — Javier Garza, Snyk
Recommendation · MoneyWhenever an agent tells you something, ask it 'how did you come up with this answer?' to surface hallucinations. For important tasks, ask the AI for four different ways of doing it and pick the one you think is better.
AI Engineer · AI Security Engineer Foundations + Certificate — Javier Garza, Snyk
Recommendation · MoneyThe creator recommends finding a better way to allow the good agents in and block the bad agents, noting that today's sign-up flows assume a human can read the landing page, fill out a form with email and password, solve a CAPTCHA, and verify their email by clicking something — flows agents can't do. An agent signing up probably doesn't have an email address, and a human can choose a plan, copy the API key out, paste it somewhere, and use it elsewhere — flows agents can't do. A dashboard experience is not really agent-native; it's a human interface, not an agent interface.
AI Engineer · https://www.youtube.com/watch?v=nSkmXakJ6eo
Recommendation · MoneyAsk the agent to ask you clarification questions when you give it an instruction; if the agent always asks about ambiguous parts of your prompt, its reliability will be better. The clarification prompt in the paper is simple: minimize ambiguity while preserving the nature of the task.
publisher-reported reliability improvement with vs without clarification
Also discussed by Tarek on Latent.Space, Thariq Shihipar on Latent.Space (unclassified source — not counted until classified). Not counted as independent sources.
AI Engineer · The Autonomous Computer: Infrastructure for Computer Use Agents — Ang Li, Simular
Recommendation · MoneyUse a symbolic layer — a coding agent — to solve the reliability problem of VLM agents. Claude Code, for example, runs the code, recognizes errors, and tries to revise the code; with auto-approval enabled, this error-recognizing revision loop is pretty much already self-improving computerization.
endorsed approach by the speaker
AI Engineer · The Autonomous Computer: Infrastructure for Computer Use Agents — Ang Li, Simular
Recommendation · MoneyThor Schaeff recommends leveraging partner solutions like LifeKit and PipeCat when scaling real-time media applications, to allow getting up and running quickly with the Gemini Live API. He also recommends visiting the partner booths at the expo.
AI Engineer · Can Your Agent Hear You Now? Building Live Voice Agents with Gemini — Thor Schaeff
Recommendation · MoneyThor Schaeff recommends trying the Live Translate model in Google AI Studio as an additional capability to explore, noting that Gemini Live Translate is available on the Gemini API and can be added to your applications.
AI Engineer · Can Your Agent Hear You Now? Building Live Voice Agents with Gemini — Thor Schaeff
Recommendation · SocialCoach Kyle advises that one of the easiest ways to practice social confidence is to voice the thoughts you already have — like wondering about a menu item or what time a place closes — out loud to strangers instead of keeping them to yourself. He frames these internal observations as natural conversation starters.
Coach Kyle · How to make SMALL TALK so natural, strangers start talking to you
RecommendationFuad Ali recommends using the detect step to surface fallback failures before a user had reported them.
AI Engineer · Building Self-Learning Loops for Your Agent — Fuad Ali, Arize AI
RecommendationFuad Ali recommends scoring the dev agent's responses with the same judge rubric you have in production.
AI Engineer · Building Self-Learning Loops for Your Agent — Fuad Ali, Arize AI
RecommendationFuad Ali recommends comparing side by side what the dev agent with the fix does versus what the production agent did, to see whether you're catching the edge cases.
AI Engineer · Building Self-Learning Loops for Your Agent — Fuad Ali, Arize AI
RecommendationFuad Ali recommends marrying trace data to an investigation of the relevant code path (e.g. the code that checks the price filter) and then proposing a fix.
AI Engineer · Building Self-Learning Loops for Your Agent — Fuad Ali, Arize AI
RecommendationFuad Ali recommends verifying with harness-as-a-judge in plain language: no column mapping or boilerplate eval templates, with the rubric being just a sentence that reads the actual traces directly.
AI Engineer · Building Self-Learning Loops for Your Agent — Fuad Ali, Arize AI
RecommendationFuad Ali recommends using agent experiments to test fixes in an automated way via CI instead of deploying to a dev environment and manually testing.
AI Engineer · Building Self-Learning Loops for Your Agent — Fuad Ali, Arize AI
RecommendationFuad Ali recommends standing up the dev agent behind an endpoint and hitting it with a dataset directly from Arize, where the dataset is actual traces your agent failed on.
AI Engineer · Building Self-Learning Loops for Your Agent — Fuad Ali, Arize AI
RecommendationFuad Ali recommends using Signal in parallel for when you don't want to be hands-on with the keyboard and want to run monitoring in the background.
AI Engineer · Building Self-Learning Loops for Your Agent — Fuad Ali, Arize AI
RecommendationFuad Ali recommends verifying with harness-as-a-judge, and then measuring with agent experiments.
AI Engineer · Building Self-Learning Loops for Your Agent — Fuad Ali, Arize AI
RecommendationFuad Ali recommends that in dataset experimentation workflows you look before the fix and after the fix: at your judge (e.g. the price correctness judge), at your examples and queries, and at whether latency or cost is increasing.
AI Engineer · Building Self-Learning Loops for Your Agent — Fuad Ali, Arize AI