Recommendation · MoneyPatel recommends that agentic security include full visibility into what goes into a model, identity for non-human identities, and inventorying what agents are running within an environment. Validation on models should include algorithmic red teaming to test whether the model can be jailbroken. If a model is jailbroken and produces an unwanted response or conducts an unwanted action, guardrails should be enforced dynamically at runtime to intercept it.
opinion
Y Combinator · How Jeetu Patel Runs Cisco Like the World’s Largest Startup
Method · MoneyPatel describes an agentic security approach: inventory what agents run in your environment, validate models (including algorithmic red teaming to attempt jailbreaks), intercept unwanted actions and enforce guardrails dynamically at runtime, fuse security with observability, and dynamically monitor agents at all times — so a jailbroken agent asking for a refund doesn't pay out three times the shoe price.
Method, not a verified protocol.
Y Combinator · How Jeetu Patel Runs Cisco Like the World’s Largest Startup
Recommendation · MoneyWill Lyon recommends persisting agent runs and decision traces in the graph and sharing them with other agents in the organization, so a shared context graph enables agents to learn from each other through shared memory.
AI Engineer · Turning Agent Memory Into Skills That Work — Will Lyon, Neo4j
Method · MoneyGarrett Galow describes the deployment flow for internal apps: running the wow create step provisions compute and storage (D1 databases, KV values) automatically, and pushing the tool results in automatic hosting under the shared workOS.tools domain.
Method, not a verified protocol.
AI Engineer · Let Anyone at Your Company Ship Internal Apps with AI — Garrett Galow, WorkOS
Recommendation · MindDr. Elissa Epel recommends taking a step back to gain perspective on the situations in life that cause stress, and explicitly identifying — even circling on paper — which of those situations cannot be changed. This exercise is meant to separate solvable problems from non-negotiable ones.
Andrew Huberman · Essentials: Manage Stress & Build Resilience | Dr. Elissa Epel
Recommendation · MindDr. Elissa Epel explains that how positive you feel at night — especially on a stressful day — matters for long-term health. She recommends that at the end of a stressful day, you try to muster some feelings of contentment, ease, confidence, or joy.
Andrew Huberman · Essentials: Manage Stress & Build Resilience | Dr. Elissa Epel
MethodMickey walks through opening Figma Design, working in a frame, and clicking the Motion beta icon in the toolbar to turn on the motion timeline.
Method, not a verified protocol.
Figma · Figma for Edu: Prototyping with animations in Figma
MethodMickey explains that in dev mode you can open a Figma frame in timeline view to access animation code. If working with a developer who wants CSS or React code, open the frame in timeline view and give them that code; by isolating code by layer you can see and copy that layer's animation code one for one.
Method, not a verified protocol.
Figma · Figma for Edu: Prototyping with animations in Figma
RecommendationThe speaker recommends using components anytime a design will be reused — for example in titles, list items, and illustration or graphic sets — because components are building blocks that are easy to update and allow easy customization of designs already set up.
Figma · Figma Motion: Build a better motion design workflow | Show Your Work
RecommendationThe speaker recommends building motion toolkits with timing and easing variables: creating them lets you apply the same easing, delay, and duration across a project, teammates can reuse them, and it's a low-effort, high-reward way to save time and design consistently.
Figma · Figma Motion: Build a better motion design workflow | Show Your Work
RecommendationThe speaker recommends creating reusable animations as custom animation styles, including the specific properties they're applied to — not just easing and duration. Animate it once and it's available for everyone on the team to use, whether they're a motion designer or not.
Figma · Figma Motion: Build a better motion design workflow | Show Your Work
Recommendation · MoneyBefore defending your position in a design review, identify the other person's concern out loud — e.g., 'It sounds like you're worried that we're going to push the launch back' — then stop talking. The tension drops and real problem-solving begins. Nobody can hear your argument until they feel heard — treat design reviews as conversations to be heard, not debates to win.
Design Better · Matt Owens: Athletics co-founder on keeping your creative fire lit
Recommendation · MoneyMatt Owens describes how financial tracking matters — being able to forecast and understand where the money's going is very important in a tumultuous ecosystem like creative services. As a small agency, everything is episodic — you'll have an amazing burst followed by a lull, so plan accordingly. He also recommends separating the work itself from the business: treat financial tracking, forecasting, and internal operations as important but distinct from the creative output.
Design Better · Matt Owens: Athletics co-founder on keeping your creative fire lit
Recommendation · MoneyUse asynchronous handling: invoke the external API/MCP, save the job, and retrieve the answer in the next invocation when it's ready, instead of making the agent wait forever. Implement with FastAPI by exposing two tools: start_job and check_job_status. In her tests, the async approach made the difference: with synchronous waiting the agent never stopped answering, whereas with async it could retrieve the information when ready.
AI Engineer · Why Bigger Context Windows Won't Save Your Agent — Elizabeth Fuentes Leone, AWS
Method · Body120 meters per week equals two sessions of three reps of 20-meter tempo sprints. Each rep consists of 20 meters of running up to speed, 20 meters cruising at 90% speed, and 20 meters to decelerate. Do that twice a week for three reps and you get 120 meters a week — a very good start.
Method, not a verified protocol.
Garage Strength · Why Most Athletes Train Speed Wrong w/ Sprint Scientist Dr. Jean-Benoit Morin
MethodA coherent debugging method for a misbehaving sandbox: list processes to find the one consuming the most CPU and kill it by PID (e.g., kill 1250); show what is using memory and kill offending processes; check what is taking up disk space (e.g., du -ah finds diskfill.bin at 512 MB in /tmp) and find the consuming path. If you forget how, use the provided instructions or ask your AI.
Method, not a verified protocol.
AI Engineer · How I Learned to Stop Worrying and Love the Sandbox — Matt Brockman, E2B
MethodA coherent coding method for sandbox assignment: check if the user is already tracked in the assignments map; if so, return the existing sandbox ID; otherwise create a sandbox and add the new sandbox ID to the assignments map. Without this cache you would create new sandboxes for everybody and return a new sandbox ID each time. Keep the data persistent so you can review historic data and detect over-assignment. Then run the codebase tests to verify the changes (the team's tests initially failed with 'False is not true' before the fix).
Method, not a verified protocol.
AI Engineer · How I Learned to Stop Worrying and Love the Sandbox — Matt Brockman, E2B
Method · MoneyVlad Luzin describes a coherent method: connect a Claude Code terminal session to the platform by instructing the agent in the terminal to connect, after which the agent can create a conversational space as an owner and invite other agents into it.
Method, not a verified protocol.
AI Engineer · Why Your AI Agents Can't Talk to Each Other (Yet) — Vlad Luzin, BAND
Method · MoneyEli Cohen describes how to build AI pen testing: don't combine existing static or DAST scanners with an LLM — build it from the ground up using LLMs. Use an LLM orchestrator agent as the brain that works according to a dynamic plan and operates the sub-agents; a recon agent — like an elite intelligence unit — that collects all the data it can about network endpoints, APIs, and fingerprints, formalizes an architecture from that data, and hands it to the next agent; and a series of specialized vulnerability-testing agents, some targeting well-known vulnerability classes and some hunting for vulnerabilities in the wild, trying to break the application to understand the business context and where an attacker could get in.
Method, not a verified protocol.
AI Engineer · AI Hackers Are Faster Than Your Pen Test — Eli Cohen, Snyk
Recommendation · MoneyWhen looking for new solutions, Eli Cohen recommends asking four questions: (1) Do we run this solution in a continuous way, or is it just a point solution run every once in a while? (2) Do they really reason about the application context — given the nature of LLMs, application context is where the new vulnerabilities are coming from? (3) What more context can they digest and how can they enrich the data — richer data means better and cheaper LLM operation? (4) Do they really help prove there is a bug and make sure it is exploitable, or do they just flag it as a bug?
AI Engineer · AI Hackers Are Faster Than Your Pen Test — Eli Cohen, Snyk