Creator perspective · MoneyAfter adding concurrency and prefetching, data transport costs still linger and have gone up. Every sample transfers large image arrays across process boundaries multiple times: pickled on the worker, copied to the coordinator, then re-pickled onto the trainer. Multimodal tensors are huge (megabytes per sample), and by default every sample is returned by value, so expensive arrays are copied from process to process. As a result the data generator can often OOM because of repeated mem copies of large multimodal tensors.
AI Engineer · From 15% to 90% GPU Utilization: Fix the Data Pipeline, Not the Model
Report · MoneyPedro Lopez describes the anti-pattern of CLIs waiting for interactive prompts: with no human present, agents get stuck.
AI Engineer · Designing CLIs for Agents, Not Humans — Pedro Lopez, Airbyte
Creator perspective · MoneyAnders Swanson argues against common memory shortcuts: relying on scratchpads (markdown files, file hierarchies, wikis, prompts) leads to ungoverned recall and does not scale in large multi-tenant environments; relying on grep for memory retrieval scales poorly and lacks complex hybrid retrieval capability; and using only larger context windows leads to token waste, compaction, and loss of useful signal in irrelevant information.
endorsed
AI Engineer · From Context to Memory: Your Agents Need a Real Memory Layer — Anders Swanson, Oracle
ReportTo train a network you must define what a wrong prediction means — an error or loss function: the closer the prediction to the actual value, the lower the loss. The empirical loss is the mean of individual losses across the dataset, and training minimizes this loss on the available data. For binary classification (e.g., pass/fail) use softmax cross entropy; for regression predicting a real number (e.g., a grade) use a loss like mean squared error.
Alexander Amini · MIT Introduction to Deep Learning (2022) | 6.S191
Creator perspective · MoneyUsers show a clear trajectory toward fully automated PRs as they build trust in the AI's accuracy: they first get reviews and build trust in the accuracy of the reviews, then give the agent teeth by allowing it to start blocking PRs, then use autofix capabilities to automatically turn PRs green and fix issues raised by code review or CI failures.
AI Engineer · AI Writes More PRs. Who Validates Them? — Ali-Reza Adl-Tabatabai, Sonar
Principle · MoneyThere is a ton of opportunity in combining an agentic approach to validation with program analysis now that Guitar is part of Sonar and integrating with SonarQ; combining agents with taint analysis, control flow analysis, data flow analysis, and software composition analysis yields much better precision and coverage at a great price point; the combination of agentic validation plus algorithmic program analysis will be much better than either alone.
AI Engineer · AI Writes More PRs. Who Validates Them? — Ali-Reza Adl-Tabatabai, Sonar
Principle · MoneyWhen building Atlas, WorkOS maintained three principles: (1) invisible infrastructure — no one should need to know where the deployment target is, and it should be transparent and swappable (Cloudflare Workers today; Vercel or AWS later shouldn't matter); (2) secure by default — no one should access sensitive internal tools outside the company, since many contain PII and customer/sales data; (3) discoverability — when someone shares a tool link in Slack, anyone should be able to click it and view the tool.
described
AI Engineer · Let Anyone at Your Company Ship Internal Apps with AI — Garrett Galow, WorkOS
Principle · MoneyAll internal apps are automatically protected by the company's Okta SSO account — users cannot log in without Okta access. For APIs and Slackbots, token-based authentication can be enabled to provide API access. All requests are logged and auditable so a leak or security issue can be detected. Secrets for internal and third-party API access are stored via a secrets manager (WorkOS uses Doppler), so builders never push secrets into a Git repo.
described
AI Engineer · Let Anyone at Your Company Ship Internal Apps with AI — Garrett Galow, WorkOS
Principle · MoneyThe most important success metric for an internal app platform is enabling non-engineers to build and deploy apps. A marker of true success is eliminating the need for engineers to migrate or manage tools on behalf of marketing or sales. 37% of WorkOS's 60 internal apps were built without any engineers in any of the commits — totally built between a person and Claude without engineering involvement.
described
AI Engineer · Let Anyone at Your Company Ship Internal Apps with AI — Garrett Galow, WorkOS
Principle · MoneyMatt Gibiec explains that if the data provided to an agent is missing context, details, or the right permissions to pull data from tools in the environment, hallucinations result.
AI Engineer · Your AI Agent Has No Nervous System — Matt Gibiec, Dynatrace
Creator perspective · MoneyAbhi Arya states that everything they did to fix the failure modes came down to one change: making uncertainty really legible. Failures should carry information — Reducto's Extract outputs confidence scores, bounding boxes, and information about what went wrong.
AI Engineer · Why We Deleted Our MCP Server and Rebuilt It — Abhi Arya, Reducto
Report · MindDr. Elissa Epel says there are practices that help people feel ease with an uncertain future. Mindful check-ins — noticing that you are carrying around uncertainty stress — are one way. She suggests reframing uncertainty as the beauty of the mystery of life and the freedom of realizing you don't control tomorrow, and viewing things with curiosity and seeing what emerges. Her exercise: instead of an alert, forward-leaning posture of trying to predict the next second, lean back and take some slow breaths. Leaning back with slow breaths helps orient you and lets you 'face time' by letting it come to you and receiving what happens. The receptive posture is a completely different body stance than the usual daytime 'go mode'. She suggests declaring internally: 'I am in a receptive mode and I'm I
Andrew Huberman · Essentials: Manage Stress & Build Resilience | Dr. Elissa Epel
Report · MindDr. Elissa Epel explains that challenge-type thoughts ('I can do this, this is a great challenge and opportunity, I have what it takes') generate a healthier hemodynamic response: more cardiac output, better oxygenation to the brain, better problem solving. People who maintain a positive outlook and challenge response are able to maintain that positive outlook through the stressor. In lab studies, people showing the challenge response are less inflamed afterward — within an hour or two they didn't trigger the pro-inflammatory response. People with the challenge response tend to have longer telomeres, suggesting a slower speed of aging.
Andrew Huberman · Essentials: Manage Stress & Build Resilience | Dr. Elissa Epel
Report · MindAndrew Huberman offers the exercise analogy: during a hard workout blood pressure, heart rate, stress hormones, and cortisol are way up, yet that sets in motion adaptations that bring you to a better place. Dr. Elissa Epel adds that short, repeated physical stressors like high-intensity interval training promote not just aerobic fitness but stress fitness. People who do short repeated physical stressors feel less rumination, less depression, and less anxiety — they tune up the nervous system. Repeated engagement with moderate stressors builds stress resilience when they are manageable — you become less threatened by them.
Andrew Huberman · Essentials: Manage Stress & Build Resilience | Dr. Elissa Epel
Principle · MindDr. Elissa Epel describes a metaphor of pulling on a rope attached to a brick wall: trying to change the unchangeable only chafes your hands while the wall never moves. When you catch yourself trying to solve unsolvable problems, she recommends telling yourself 'drop the rope.' Dropping the rope frees your hands so you can live in the ways you do have control over, even though the brick wall remains.
Andrew Huberman · Essentials: Manage Stress & Build Resilience | Dr. Elissa Epel
PrincipleFedus explains why RL directly on live experiments (kick off an experiment, wait days, update on whether you found a room-temperature superconductor) is infeasible: rollout time, instrument queues, and furnace physics make variance reduction impossible with available agents. Instead, Periodic Labs constructs agents around the dataset the lab has already produced.
Latent.Space · Synthesis Superintelligence: from Semiconductors to Superconductors — Periodic Labs’ Liam Fedus and Ekin Dogus Cubuk
Creator perspectiveThe guests describe the materials discovery loop as starting with figuring out what to make: assessing the stability of the atoms together and whether the material is expected to have the desired properties. The next step is figuring out how to synthesize it, which is non-trivial — even when a configuration is known to hold together, determining the processing conditions to realize it is highly non-trivial. They stress that a novel configuration of atoms is not the goal; the atoms must exhibit the property of interest when put together.
described as the lab's working model of discovery
Latent.Space · Synthesis Superintelligence: from Semiconductors to Superconductors — Periodic Labs’ Liam Fedus and Ekin Dogus Cubuk
ReportFedus describes the materials discovery loop in three stages: (1) decide what to make — a configuration of atoms that is stable and exhibits the desired property (not merely novel); (2) figure out how to synthesize it — processing conditions; (3) characterize what you actually made, since products don't come out labeled.
Latent.Space · Synthesis Superintelligence: from Semiconductors to Superconductors — Periodic Labs’ Liam Fedus and Ekin Dogus Cubuk
Principle · MoneyNan Yu says the most successful efforts map agent design onto human nature, because people evolved to manage a limited number of parallel threads; managing 40 agents is too many for most people to keep up with — ask whether you could manage 40 direct employees.
Lenny's Podcast · Stop planning for 2027: how OpenAI builds product 90 days at a time | Tara Sesha and Nan Yu (OpenAI)
Creator perspective · BodyRunning backs are taught to run downhill (low and forward), which tends to give them higher backside mechanics; the objective when working with them is that they remain good football players, not that their running technique be changed. You should not make position players upright just for the sake of protecting hamstrings, because it would be detrimental to their game.
Garage Strength · Why Most Athletes Train Speed Wrong w/ Sprint Scientist Dr. Jean-Benoit Morin