Recommendation · MoneyWhen a tool needs external resources, the MCP server requests a token from the STS for the downstream resource. The STS evaluates policy, the delegation chain, and requested access, then mints a resource token scoped to that tool call, keeping the user subject and adding the app that performed the exchange to extend the delegation chain for activity tracking.
AI Engineer · Your Agent Is a Lethal Trifecta (And What to Do About It) — Kim Maida, Keycard
Recommendation · MoneyLow-risk activities like assessing a ticket might be auto-approved under admin policy, while higher-risk missions like remediating a usage spike are presented to the user for review and approval before the agent acts.
AI Engineer · Your Agent Is a Lethal Trifecta (And What to Do About It) — Kim Maida, Keycard
Recommendation · MoneyGive the on-call agent an MCP server that provides tools and also brokers access to resources; a custom MCP server can also exchange tokens for downstream resources, and it should provide a defined set of production infrastructure tools.
AI Engineer · Your Agent Is a Lethal Trifecta (And What to Do About It) — Kim Maida, Keycard
Recommendation · MoneyIn the build execution template, Nnenna Ndukwe recommends documenting the starting state — what branch you're in and what the current local verification process is — plus any commands expected to run to determine that code passes or fails expectations with linters, and the implementation skill handoff and execution steps. She wants to know that the agent knows every single step in the process of making the code change and determines what the test plan is.
AI Engineer · How to Build Quality Gates into Agentic Coding Workflows — Nnenna Ndukwe, Qodo AI
Recommendation · MoneyNnenna Ndukwe demonstrates using the repo-local workshop planning skill to turn the hands-on task in the README into a high-level implementation plan and a build session execution plan. The planning prompt grounds the plan in agents.md, README, and the rules, and includes verifying the local Kodo setup and dependency health before selecting rules. In planning mode, the agent only reads and inspects files — no code or plan files are written in that mode.
AI Engineer · How to Build Quality Gates into Agentic Coding Workflows — Nnenna Ndukwe, Qodo AI
Recommendation · MoneyNnenna Ndukwe recommends getting your agent ready before it starts on a project: copy-paste the agent prompts from the workshop README, ask the agent to install or verify Kodo skills, read agents.md and rules, and select the relevant repo-local rule IDs for the task. Optionally compare those rules with Kodo get rules if those skills are available. If the agent doesn't prompt you to choose which task you're doing, make that clear as part of the prompt.
AI Engineer · How to Build Quality Gates into Agentic Coding Workflows — Nnenna Ndukwe, Qodo AI
Recommendation · MoneyNnenna Ndukwe recommends including skill routing in the plan to plan ahead all the skills you would expect your coding agent to kick off one by one during implementation. This is especially helpful if you want to use loops or goals in Codex or Claude Code to work on the task without stepping in. The plan must include an implementation skill handoff that names the first implementation skill and the test-driven development prompt that will be chained to and run next.
AI Engineer · How to Build Quality Gates into Agentic Coding Workflows — Nnenna Ndukwe, Qodo AI
Recommendation · MoneyJavier Garza describes free Snyk tools: the Snyk free MCP scanning agent, run by typing a UBX command in the terminal, which installs and runs it; SNCC agent scan to scan installed MCP tools for tool poisoning, toxic flows, prompt injection, and CI checks; SNCC code scan to scan downloaded MCP code from GitHub for insecure libraries, path traversal, command injections, and SQL injections; the SNCC CLI to automatically fix insecure MCP server code, not just detect vulnerabilities; and the SNCC extension for Visual Studio Code to identify path traversal vulnerabilities with line numbers and fix guidance.
AI Engineer · AI Security Engineer Foundations + Certificate — Javier Garza, Snyk
Recommendation · MoneyJavier Garza recommends keeping AI-native IDEs and CLIs (Claude Code, Cursor, Windsurf) updated to the latest version, though he sometimes recommends waiting a day or a few hours after a new release, since many updates fix issues or memory leaks.
speaker hedges on timing
AI Engineer · AI Security Engineer Foundations + Certificate — Javier Garza, Snyk
Recommendation · MoneySet a time bound on eval runs because loops can stall, hardware can fail, or bad agent code can hang the entire system. In their case, the bound was 210 seconds: 60 seconds for training, 60 seconds for testing on unseen text, and a 90-second buffer for GPU spin-up and overheads.
AI Engineer · Parameter Golf with AutoResearch — Vayum Arora, Zhengyao Jiang, Dixing Xu & Dhruv Srikanth, Weco AI
Recommendation · MoneyIt's difficult to know a priori how many steps you should give your AutoResearch system; Weco chose 20 steps for the workshop to give agents enough time to explore different ideas and see what the agent is exploring, what works, and what doesn't. More ideas that get tried generally leads to more breakthroughs, and if you have a fast or cheap eval, then under a time or cost constraint you can run more steps. You also want to bound the time for which your eval runs, because a lot of things can go wrong: loops can stall, hardware can fail, and sometimes the agent produces really bad code that causes the entire system to hang. In their case, they set the eval timeout to 210 seconds: 60 seconds for training the model, 60 seconds for testing the model on unseen text, and 90 seconds as a buffer.
AI Engineer · Parameter Golf with AutoResearch — Vayum Arora, Zhengyao Jiang, Dixing Xu & Dhruv Srikanth, Weco AI
Recommendation · MoneyConcrete tricks to encode in skills: encode the whole query in the URL as the strongly preferred fast path; add explicit waits when a website loads slowly; don't retry blindly on open errors — follow the learned skill steps instead; persist learnings like 'go straight to the schedule page', 'type the keyword into the search input', 'read the results'.
AI Engineer · Hill-Climbing Skills: Improve Agents Without Changing the Model — Shubhankar Srivastava, Browserbase
Recommendation · MoneyShubhankar Srivastava recommends protecting against prompt injection at the infrastructure layer using Browserbase safe mode to control which URLs the agent can visit and what actions it can take. You can also implement action permission modes like Codex or FlotCode — always allow, allow once, or allow any action — using Browserbase safe mode.
AI Engineer · Hill-Climbing Skills: Improve Agents Without Changing the Model — Shubhankar Srivastava, Browserbase
Recommendation · MoneyDakkak recommends using Modular's AI skills tooling to translate code from languages developers already know into Mojo or Max, and warns that rewriting from scratch wastes time when porting tools exist. He also suggests testing the tools on a small module before a full migration.
Advice from the speaker; effectiveness of the tooling is not independently verified.
AI Engineer · Taming the AI Hardware Cambrian Explosion — Abdul Dakkak, Modular
Recommendation · MoneyJonathan says he has been pushing his team to use Jam MCP as the standard way to capture UI/UX details and acceptance criteria from non-technical people, including recordings that capture console logs and network requests that can be fed into an AI harness.
AI Engineer · Build with Perception Agents — Emile Baizel & Shruti Arora, Amazon AGI Lab
Recommendation · MoneyDan recommends measuring accuracy including hallucination rate — 3% in his example — determined partly automatically and partly via manual annotation.
AI Engineer · https://www.youtube.com/watch?v=PwCSYy-oGsM
Recommendation · MoneyUse the MCP search engine API first to find target pages, then call the data pipeline API to pull structured results — a pattern illustrated by a client building a social media listener who needed follower counts across Instagram, TikTok, and LinkedIn.
stated
AI Engineer · https://www.youtube.com/watch?v=IK8cC4FScps
Recommendation · MoneyWhen building a resume-matching model, source jobs data that includes which companies are posting, how many jobs are available, salary, and location — and ensure data flows to you consistently because performance depends on continuous delivery.
stated
AI Engineer · https://www.youtube.com/watch?v=IK8cC4FScps
Recommendation · MoneyThor says the easiest way to try the Gemini Live API is through Google AI Studio: go to ai.studio/live (or ai.dev/live) to talk to the model and explore its capabilities. In AI Studio you can select Gemini 3.1 Flash Live, choose a different voice (e.g., Puck), and enable grounding with Google Search. The Live Translate model can also be tried there.
AI Engineer · Can Your Agent Hear You Now? Building Live Voice Agents with Gemini — Thor Schaeff
RecommendationRun custom harness judges in CI on agent-proposed PRs, experiment with them, and optimize from there.
endorsed
AI Engineer · Building Self-Learning Loops for Your Agent — Fuad Ali, Arize AI