Report · MoneyA production agent at thousands of requests per second produces millions and millions of traces, and humans become a bottleneck: you cannot get an overall sense of what all your traces are doing by reading them one at a time.
AI Engineer · From Vibes to Production: Evaluating and Shipping AI Agents That Work 201 — Laurie Voss, Arize AI
Creator perspective · MoneyAn agent loop is a model with tools in a loop — think, act, observe, decide what to do next — often going back through itself repeatedly. Claude, Codex, and others ship built-in helpers for loops, and you can build your own loop workflow that works exactly for you, your team, or your company.
AI Engineer · Lifestyles of the AI-Native — Nick Nisi & Zack Proser, WorkOS
Report · MoneyPolicies — internal and external — change a lot over time, so your eval pipeline must account for that. At a large company, policy becomes living — over decades, policies change and things become different. Ideally, you want living policy that an agent can discover automatically, rather than writing down the rules manually. Agentic RAG is part of how you implement living policy.
AI Engineer · Evals in AI: A Deep Dive — Tejas Kumar, IBM
Report · MoneyJust because your eval pipeline is green doesn't mean it works — be sensitive to these biases as you build your eval pipeline. Evals have position bias, sycophancy, self-preference, and verbosity bias; a green pipeline can still be lying to you.
AI Engineer · Evals in AI: A Deep Dive — Tejas Kumar, IBM
Report · MoneyUnit tests are reliability instrumentation for deterministic systems (expect(add(1,2)).toBe(3) holds every time), but with AI almost nothing is deterministic, so unit tests can no longer be used for AI reliability. Evals are reliability instrumentation for non-deterministic systems — like tests, but fuzzier. Unit tests ask 'did this function return exactly this number?' while evals ask 'did the agent support the scenario?' — evals deal in scenarios.
AI Engineer · Evals in AI: A Deep Dive — Tejas Kumar, IBM
Principle · MoneyThe fundamental job difference: an account executive runs one five-stage sales cycle selling the promise of value, while an account manager must deliver the proof that the promise was kept before earning another cycle. AEs can drop an account and cold call another; AMs cannot — their prospecting is value delivery.
30 Minutes to President's Club | No-Nonsense Sales · #604 - Upsell & Expansion Masterclass: 7 Lever to Grow Any Account
ReportIvan Leo frames the Interactions API (riffing on 'Attention is all you need') as Google's solution to the difficulty of consuming models and agents through many different endpoints, and states that all of Google's new models will be on it.
AI Engineer · An Interaction Is All You Need — Ivan Leo, Google DeepMind
ReportAnti-gravity has launched as a remote agent — what Google calls a managed agent — standardized across the Google suite of products. In AI Studio, websites are built with the anti-gravity agent; in the anti-gravity application you're also using the anti-gravity agent. The anti-gravity agent is fundamentally the same agent running in the anti-gravity IDE, so locally-tuned workflows and skills transfer to the cloud.
AI Engineer · An Interaction Is All You Need — Ivan Leo, Google DeepMind
Creator perspective · MoneyHarald Kirschner explains you don't care about individual errors; you just want a more stable code base. Run the error pipeline as a mix of agents and deterministic systems: automated triage, then filing, then handoff to an agent for the PR in a multi-agent system. Fixes land mostly automatically with some human oversight to still approve the PR.
AI Engineer · How VS Code Went from Monthly to Weekly Releases with AI — Harald Kirschner
Principle · MoneyThe core difficulty of browser agents at scale is that each step incurs incremental cost and incremental failure risk, while value arrives only when all steps complete — there is no partial credit, and the vast majority of web automations fall into this category.
: none
AI Engineer · Why 99% Accurate Browser Agents Still Fail — Derek Meegan, Browserbase
Principle · MoneyDerek Meegan contrasts a one-off browser agent demo (one site, one run, a good day, watched, fingers crossed) with production deployments: thousands of unattended runs while the site can change or the model can have an off day.
: none
AI Engineer · Why 99% Accurate Browser Agents Still Fail — Derek Meegan, Browserbase
Report · MoneyHyperscalers (Alphabet, Amazon, Meta, Microsoft, Oracle) are investing nearly all near-term operating cash flow into capacity due to insatiable demand. This creates a J-curve where free cash flow is depressed during buildout but is expected to recover from 2028 as GPUs/TPUs have long useful economic lives.
described
The a16z Show · The $1 Trillion AI Buildout | State of Markets
Creator perspective · MoneyAlex Atallah counters the view that distribution platforms are 'thin wrappers', arguing that orchestrating multiple APIs requires significant engineering and community design.
Latent.Space · OpenRouter: from Seed to Stripe — with OpenRouter’s Alex Atallah & AMP’s Anjney Midha
Creator perspectiveThe host describes the core principle of magazine making as the tension between image and text, where putting text on an image changes its meaning entirely.
Figma · 8 Designs That Inspired i-D Magazine's Editor in Chief | Source Material
Creator perspectiveThe host states that print is a way to counter people's fatigue with the digital world, and that they like to think about print in the most tactile terms possible.
Figma · 8 Designs That Inspired i-D Magazine's Editor in Chief | Source Material
Creator perspectiveWorld models are a new category of AI model, analogous to language models, but based in visual and physical understanding. They are designed to generate, simulate, and reconstruct worlds, and are intended as a general horizontal platform applicable to many industries.
The a16z Show · World Models, Robotics, and the Future of 3D AI
ReportOfir Ehrlich and Gonen Stein explain that data within large organizations is frequently locked away by business unit owners with different incentives. Data teams struggle to access it without compromising security, compliance, or production uptime, and often don't even know what data exists.
No Priors: Artificial Intelligence | Technology | Startups · Rethinking Legacy Data Infrastructure with Eon Co-Founders Ofir Ehrlich and Gonen Stein
Recommendation · MoneyIn the recorded run chain, clearly delineate deterministic steps (code-based fetch or match) from agent steps (extraction, conclusion).
AI Engineer · Your Financial Agent Needs to Show Its Work - Anirudh Veeraragavan, Fieldguide
Recommendation · MoneyRerun individual recorded steps rather than the whole run to isolate exactly where a result diverged between points in time.
AI Engineer · Your Financial Agent Needs to Show Its Work - Anirudh Veeraragavan, Fieldguide
Recommendation · MoneyUse Skill Doctor, a skill you can run on your single-player setup that crawls through all of your agent conversations in Claude, Codex, Pyre, whatever else, and does a one-time grading. If you haven't done this grading approach before, run it across your setup and see what happens — it's a first go to see what you can optimize right now and get some low-hanging fruit. It grades in batches for a little bit of token efficiency and produces a skill report card showing how the agent performed across all defined rubrics. While grading should probably be continuous rather than whenever someone remembers, a one-shot run is a nice starting point.
Consequential creator advice requires independent verification before Today can promote it.
AI Engineer · Your AI Agent Is Hiding Its Mistakes From You - Ben Holmes, Warp