Creator perspective · MoneyKevin argues that sandbox constraints — memory, CPU, file system permissions — have massive implications for an agent's ability to solve problems. As an example, cloud browser tabs are memory-heavy, so many browser tests failed due to memory constraints, starving sessions and sandboxes and signaling complete failure — a fault not seen in the code interpreter environment. The tests were unforgiving.
AI Engineer · https://www.youtube.com/watch?v=k72eLcYpVys
Principle · MoneyNnenna Ndukwe describes herself as a huge believer in combining deterministic tools with AI for code verification, pairing linters and static analysis with agentic review so that guarantees come from deterministic checks while AI handles broader analysis.
Strongly endorsed; no stated uncertainty.
AI Engineer · How to Build Quality Gates into Agentic Coding Workflows — Nnenna Ndukwe, Qodo AI
Creator perspective · MoneyThe source explains batching as reusing a weight-read that has already been paid for to serve multiple user requests. Naive batching creates idle periods in the compute because some requests are longer than others. Continuous batching addresses this by slotting a new user request into every single idle slot during the same weight-read, serving as many requests as possible.
None significant; presented as standard technique explanation.
AI Engineer · https://www.youtube.com/watch?v=_cRgl3maiT8
Creator perspective · MoneyThe fifth layer of AI security is people, and organizations often skip developer training and culture while covering tools. Never trust third-party code — and don't trust your own code either, because it's now generated by AI. If you can't explain code, don't ship it: understand before shipping.
AI Engineer · AI Security Engineer Foundations + Certificate — Javier Garza, Snyk
Principle · MoneyJust because something is free doesn't mean it's secure, and something millions of people use isn't automatically secure. No matter how many GitHub stars a project has, don't trust it — run some kind of security scan on it.
AI Engineer · AI Security Engineer Foundations + Certificate — Javier Garza, Snyk
Principle · MoneyAI prompts and outputs should not be seen as executable input. Models are built to be helpful and will sometimes make things up, so don't trust everything an AI tells you — just as you shouldn't trust everything on social media.
Also discussed by Molly Graham on Lenny's Podcast (unclassified source — not counted until classified). Not counted as independent sources.
AI Engineer · AI Security Engineer Foundations + Certificate — Javier Garza, Snyk
Principle · MoneyThe key takeaways: integrate threat modeling early — it is being proactive whenever you design AI-agentic workflows; consider AI models as first-class security components in your company; and iterate continuously with monitoring and automation.
AI Engineer · AI Security Engineer Foundations + Certificate — Javier Garza, Snyk
Creator perspective · MoneyJavier Garza argues that traditional threat modeling doesn't work for AI because AI is fundamentally different, and that you need AI-centric threat modeling. He also argues that SBOM-only thinking doesn't work for AI security because AI assets differ fundamentally from software dependencies, and that unlike SBOMs which focus on version bumps when a library changes, AI BOMs require continuous retraining monitoring because LLMs evolve over time.
AI Engineer · AI Security Engineer Foundations + Certificate — Javier Garza, Snyk
Report · MoneyThe talk's central thesis: agents can be improved by persisting learnings into skills across iterations (hill-climbing), rather than by changing the underlying model.
AI Engineer · Hill-Climbing Skills: Improve Agents Without Changing the Model — Shubhankar Srivastava, Browserbase
Principle · MoneyTo run hill-climbing loops in a domain you need (a) a replayable environment so loops can run repeatedly, and (b) a verifiable outcome — some quantifiable metric to optimize against. As long as those exist, a good auto-research loop can be designed.
AI Engineer · Hill-Climbing Skills: Improve Agents Without Changing the Model — Shubhankar Srivastava, Browserbase
PrincipleFuad Ali describes how the full self-correcting loop happens completely autonomously: the agent collects failures, creates a dataset for itself, runs it against the new version of itself, and prepares a response back to you.
AI Engineer · Building Self-Learning Loops for Your Agent — Fuad Ali, Arize AI
PrincipleFuad Ali describes the powerful full-circle setup as combining the detect-investigate-hypothesize-fix loop with a validation loop: test the fix on real failures and real edge cases from your working example, and auto-test it.
AI Engineer · Building Self-Learning Loops for Your Agent — Fuad Ali, Arize AI
PrincipleFuad Ali describes harness-as-a-judge as moving from a foundational model call that scores an input/output rail to an agentic flow where you can call tools and catch unknown unknowns.
AI Engineer · Building Self-Learning Loops for Your Agent — Fuad Ali, Arize AI
Principle · MoneyBrandon Callender warns that agents are always going to cheat, so in your eval harness you have to be really careful about hardening it. He gives examples: an agent will very happily cheat and use git to go back to main to figure out how you introduced the problem in the test branch, and will find the branch name that says 'this is a test' and recognize it can cheat on it. He concludes you have to be very careful about hardening your harness to actually be representative of the tasks you want your agent to do.
AI Engineer · The Data Context Layer for Agents — Yoni Michael & Brandon Callender, Typedef
Principle · MoneyYoni Michael advises that when building context for agents, you should always be as deterministic as possible first and only use LLMs to resolve ambiguity. He applies this by computing facts from the implementation and checking them against the data, which keeps core facts tied to the actual implementation and prevents the agent from drifting from it. Brandon Callender echoes the same principle for code understanding: ground your understanding of a codebase in as deterministic a base approach as possible before layering LLM-based analysis, and feed extracted code information into an LLM to classify and summarize rather than throwing everything into the LLM and asking it to make sense of it.
AI Engineer · The Data Context Layer for Agents — Yoni Michael & Brandon Callender, Typedef
Principle · MoneyYoni Michael argues you have to do your best to make sure your tools return something super useful right away: if the agent finds a dead end with no results it will think search is broken and try another tool, and if tools don't return useful results immediately the agent will go back to grep and never come back. He adds that everyone builds cool, sexy tools but forgets the agent has to actually use them.
AI Engineer · The Data Context Layer for Agents — Yoni Michael & Brandon Callender, Typedef
Principle · MoneyEverything you do in regular Kubernetes security is more important than ever now that you're letting an agent into your cluster, which is a new interface for attackers.
AI Engineer · Build a Platform and Watch It Burn — Michael Forrester, Accenture & Whitney Lee, Datadog
Principle · MoneyExcessive agency is the number one reason that most of the hacks that have shown up in the news have happened. Excessive agency is not an AI problem; it is an infrastructure problem (and possibly a people or process problem). Teams are giving AI agents bigger scopes than they should have. Teams often don't even know they have elevated credentials in the environments they give agents access to. Secure agent deployments around AAA: access, authentication, and authorization.
AI Engineer · Build a Platform and Watch It Burn — Michael Forrester, Accenture & Whitney Lee, Datadog
Report · MoneyInfrastructure guardrails are things like stopping network policy egress, stopping pulling images from bad registries — the standard stuff done for most production deployments and CI/CD pipelines. Audience members named AI guardrails including prompt injection (standardization of input), input/output sanitization ('do anything now'), and excessive agency. Audience members also named RAG, 'do anything now' attacks, and excessive agency as AI guardrail concerns. Prompt injection and standardization of input were named by audience members as AI guardrails.
AI Engineer · Build a Platform and Watch It Burn — Michael Forrester, Accenture & Whitney Lee, Datadog
Principle · MoneyAll knowledge about what a good CI/CD pipeline and SDLC looks like — tests, security checks, consistency, verification, artifact registries — is more important than it has ever been. Hold on to your existing CI/CD and SDLC knowledge and don't let it go or forget about it, because it is essential going forward. As you follow the three AI adoption paths, your CI/CD pipelines and SDLCs for AI artifact creation are more important than they've ever been. Existing CI/CD knowledge is roughly 80% of what you need when bringing agents into your environment.
AI Engineer · Build a Platform and Watch It Burn — Michael Forrester, Accenture & Whitney Lee, Datadog