Method · FuelSlather a little mayo on the toasted bun as glue so the dressed shrimp don't fall out. Stack in about three ounces of finished shrimp salad — don't overfill, or the shrimp-to-bun ratio gets out of whack and the shrimp start falling out, making it hard to eat (it's always a dance: it looks like it can take more, but too many overflow). Add a couple cracks of fresh pepper — not too hard, as pepper can overpower the shrimp — plus a little more chive for presentation. Old Bay would also be a fun topping, but be careful with it like the pepper. Finally squeeze the roll together so it's cohesive and the shrimp are tucked in.
Method, not a verified protocol.
Brian Lagerstrom · Skip the $40 Lobster Roll. Make This Instead.
MethodOnce you're happy with skills you've tuned locally, you can package them into a single folder and upload them the same way using the sources into a .agents folder. Uploading your skills folder gives your agent the exact same skills with the same agent harness and the same prompt running locally and in the cloud. When you're comfortable with what your agent is doing locally, you can push it up and everything works as intended.
Method, not a verified protocol.
AI Engineer · An Interaction Is All You Need — Ivan Leo, Google DeepMind
MethodTo build something similar: pass Gemini 3.1 Flash to generate an image (e.g., 'generate an image of me in Venice'), take the Interaction ID, chunk it back into client.interactions.create, and specify the Omni Flash model launched on Wednesday. To generate an image instead of audio, use the same client.interactions.create method and just modify the response modality and generation config; the output.type becomes image.
Method, not a verified protocol.
AI Engineer · An Interaction Is All You Need — Ivan Leo, Google DeepMind
MethodOnce your agent is performing well, you can configure and set it up as a named agent, much like the anti-gravity preview agent. The first way is to freeze it with a certain set of configured files and network sources. The second way is by chatting iteratively with an agent to set up an environment as you like, then freezing that environment as a specific agent so all subsequent calls go through it. The platform supports up to 1,000 named agents, allowing many different configurations. Once you're happy with your named agent configuration, it makes it very easy to scale up your workloads. You don't pay for storage or the sandbox; you only pay for the model.
Method, not a verified protocol.
AI Engineer · An Interaction Is All You Need — Ivan Leo, Google DeepMind
RecommendationAlex Zhang argues that when competing against frontier labs, you should compete on dimensions you can control rather than replicating their setup — if you use the same setup, you're only competing on data and compute, which you cannot win. Don't try to compete with frontier labs by building an autoregressive decoder-only model, since you cannot match their compute scaling resources. If you're a small AI lab, do something other than replicating OpenAI/Anthropic — anything other than the frontier-lab playbook; a startup whose strategy is just to replicate OpenAI or Anthropic has a horrible strategy.
Latent.Space · Academia is for Ambition — Alex Zhang, MIT
Recommendation · MoneyWhen pre-merge review is automated, production monitoring becomes the last reviewer standing; teams must monitor what code actually does in the wild.
medium
AI Engineer · The Death of the Code Review: What the Data Actually Says — Laurie Voss, Arize AI
Recommendation · MoneyHuman review should be retained for code where correctness is not cheaply checkable, blast radius is large, or accountability is required.
medium
AI Engineer · The Death of the Code Review: What the Data Actually Says — Laurie Voss, Arize AI
Recommendation · MoneyThe gap between general AI knowledge and company-specific context drives the opportunity for application-layer services. The current opportunity is taking capabilities and harnessing them to build reliable services, bridging the gap between model capabilities and usage.
endorsed
The a16z Show · The $1 Trillion AI Buildout | State of Markets
Recommendation · MoneyAgents should assess product quality based on intrinsic merit rather than marketing performance or brand dominance, unless the user explicitly opts into brand preference.
high
The a16z Show · The Personal Agent Race Is Here | Anish Acharya & David Pawlan
Recommendation · MoneyUse AI agents to relieve bureaucratic pressure and administrative burdens for average consumers, providing access to legal or financial optimization similar to what wealthy individuals have.
high
The a16z Show · The Personal Agent Race Is Here | Anish Acharya & David Pawlan
Recommendation · MoneyDo not implement proactive AI interventions that cross the 'fine line' of user trust; one misstep in proactivity (e.g., breaking up with someone) destroys all user trust.
high
The a16z Show · The Personal Agent Race Is Here | Anish Acharya & David Pawlan
Recommendation · MoneyIn fast-moving environments, prioritize comfort with ambiguity and strong judgment over past specific achievements. Past credentials matter less than the ability to actively collaborate and experiment in the new context.
high
The a16z Show · Building a Team at AI Speed | Harvey’s Maggie Landers
RecommendationMonitor internal APIs and services for denial-of-service style attacks from AI swarms. Nobody thinks internal GitHub or Slack is vulnerable to denial of service attacks. Swarms will look like a denial of service attack.
The a16z Show · Aaron Levie, Steven Sinofsky & Martin Casado: How Do You Secure a World of AI Agents?
RecommendationImplement granular file system permissions for AI agents rather than all-or-nothing access. The model we have is not granular enough. It is difficult to make granular security controls intuitive for users.
The a16z Show · Aaron Levie, Steven Sinofsky & Martin Casado: How Do You Secure a World of AI Agents?
MethodAnthony demonstrates using a published MakeKit by clicking 'Add' and typing a prompt, and explains how to update it by going back to the original file and clicking 'Publish Kit' then 'Update'.
Method, not a verified protocol.
Figma · Bringing Design Systems into Figma Make
RecommendationRichard Socher recommends using open-endedness methods (inspired by evolution) instead of specific rewards: set high-level environment and reward goals, then let AI agents co-adapt through interplay. He also suggests allowing an AI agent to set its own goals.
medium
Latent.Space · Humanity’s Last Invention — Richard Socher of Recursive
RecommendationRichard Socher endorses supporting open source AI, arguing it is better for security (more good actors) and for soft power competition with China. He specifically recommends supporting Western open-source language models to counterbalance Chinese soft power and propaganda.
Latent.Space · Humanity’s Last Invention — Richard Socher of Recursive
RecommendationRichard Socher argues that code is neurosymbolic reasoning and that people underestimate LLMs' ability to code. He recommends deeply integrating this capability into the architecture to unlock further growth.
Latent.Space · Humanity’s Last Invention — Richard Socher of Recursive
RecommendationRichard Socher endorses replacing manual feature engineering with learned systems (vectors and neural nets) and replacing manual architecture engineering with a unified general architecture instead of building special neural nets for each task.
Latent.Space · Humanity’s Last Invention — Richard Socher of Recursive
RecommendationRichard Socher endorses automating the AI research process itself (ideating, implementing, and validating ideas) to create a self-improving AI, and recommends reading the Darwin-Gödel machine paper by Jeff Clune for foundational ideas on recursive self-improvement.
Latent.Space · Humanity’s Last Invention — Richard Socher of Recursive