Report · MoneyMaintain evaluation criteria over time so evaluators do not over-index or become uninformative, and maintain and expand your dataset by adding good production examples to offline evals. Run experiments on those datasets and evaluate them to always have a good indication of app quality. Ship changes with confidence by making evals part of your CI/CD so changes don't regress, and deploy only when you know you can ship with confidence based on eval results. All of this is not a one-off — you need to put in the reps and close the loop continuously.
AI Engineer · Continuously Improving Agents with Langfuse — Annabell Schäfer & Lotte Verheyden, ClickHouse
Report · MoneyBecause you don't sit next to users interacting with the application remotely, you want to track what the agent is doing — and not only track it, but actually make sense of it. Get a step-by-step view into what is going on: which tools the agent uses and what the inputs and outputs are. The trace is rooted in an agent observation, with generations and tool calls below it where context is pulled.
AI Engineer · Continuously Improving Agents with Langfuse — Annabell Schäfer & Lotte Verheyden, ClickHouse
Principle · MoneyThe presenters advise designing evaluators around the typical failure modes of your application rather than copying what already exists.
AI Engineer · Continuously Improving Agents with Langfuse — Annabell Schäfer & Lotte Verheyden, ClickHouse
Report · MoneyTo test that evaluators work, go back into the IT support agent, ask the Bluetooth question, then reply with disagreement ('no, there is no such Bluetooth') and frustration ('this button is actually not there. I don't know where to find it and it's not true what you're saying') to generate new traces. Then check the traces view to verify new traces and evaluator scores are coming in, and dive deeper into an individual trace to inspect the user message, the reply, and the evaluator score. The agent replied 'okay, dad, no worries, let's try the other way first', and the disagreement evaluator scored the trace.
AI Engineer · Continuously Improving Agents with Langfuse — Annabell Schäfer & Lotte Verheyden, ClickHouse
Report · MoneyLotte Verheyden describes instrumenting the application and deploying to production so traces come in, and monitoring via automatic evaluators, dashboards, user feedback, and manual review of sample traces.
AI Engineer · Continuously Improving Agents with Langfuse — Annabell Schäfer & Lotte Verheyden, ClickHouse
Creator perspective · MoneyLotte Verheyden explains that automatic evaluators scale to all traces but are less likely to catch new or nuanced cases, while user feedback and human annotation scale less well but detect things you haven't thought of yourself; user feedback also surfaces the important traces worth looking at as the application scales.
AI Engineer · Continuously Improving Agents with Langfuse — Annabell Schäfer & Lotte Verheyden, ClickHouse
Principle · MoneyWolfram Ravenwolf says the leaderboard is not just an answer — it helps you find a better question, like what are you looking at: do you care about the capability of a model or the reliability of a model? He recommends using the leaderboard to help find a better question to ask about a model, not as a final answer, and deciding whether you care about the capability of a model or the reliability of a model before interpreting benchmark results.
endorsed
AI Engineer · From Zero to Leaderboard: Agent Evaluation — Wolfram Ravenwolf, Weights & Biases
Principle · MindIf freewheeling evolution leads in a dystopian direction, there would be high value in having some option to change track — but that would require coordination. If only one group decides to do something not economically optimal, it would just be outcompeted by other entities going for economic productivity — hence coordination is required.
The a16z Show · Nick Bostrom: How Should We Navigate Superintelligence?
Principle · MoneyJosh Woodward recommends that when a team faces an uncertain strategic debate, it can run experiments designed to prove both sides of the argument rather than resolving the debate by opinion alone.
Lenny's Podcast · Why every company now needs to think and operate like a lab team | Josh Woodward (VP Google Labs)
Principle · MoneyJosh Woodward warns that a frontier team must bridge and connect across the bigger company so it isn't dismissed as a weird R&D group making gimmicks and toys. You can get the team and environment right and still fail on the bridging — and then the whole thing doesn't matter. Someone on the team must be able to look at a toy-like technology and recognize it will be essential to the company's future in five years.
Lenny's Podcast · Why every company now needs to think and operate like a lab team | Josh Woodward (VP Google Labs)
Principle · MoneyModel routing can be set up so the tool-calling layer picks which model to use for the underlying tool based on the task, rather than being tied to the tool itself.
Also discussed by Jeff on Y Combinator (unclassified source — not counted until classified). Not counted as independent sources.
AI Engineer · https://www.youtube.com/watch?v=Ndp5VaqFrl0
Principle · MoneyIn the structured-query paradigm, the model doesn't answer the question directly; it writes the query that accesses the database. Generating the query is not the only thing required: to generate queries reliably and handle security concerns you need schema awareness, relationship understanding, security filtering, and self-correction.
AI Engineer · Beyond RAG: Build a Relational Context Engine from Scratch — Peter Werry, Unblocked
Creator perspective · MoneyThorsten endorses adopting WebAssembly on the server side as a solution to the frictions of region-bound architectures, describing it as a new way to build back-end applications that is being done successfully. He offers the analogy that WebAssembly relates to containers as containers relate to virtual machines: binaries that are super tiny, easy to distribute, super fast to run, and portable.
AI Engineer · Zero Cold Starts: Serverless AI Agents on Akamai Functions — Thorsten Hans
Creator perspective · MoneyThe creator argues that getting the workload off your machine and into CI is key for optimizing costs: without visibility into how agents work with the tool, you can't tell whether they're too verbose, achieving the task, when tool calls happen, or whether they succeed — and you have no opportunity to optimize. Once agents are in the pipeline, cost optimization, A-B testing of models, and performance analysis become possible.
AI Engineer · From Your Laptop to the Pipeline: Scaling Custom Agents with GitHub Copilot
Creator perspective · MoneyWorking locally on a single session or device means a device failure or network loss interrupts the workflow; the creator argues agents should run in the platform so they run autonomously and are not interrupted by local device or network issues.
AI Engineer · From Your Laptop to the Pipeline: Scaling Custom Agents with GitHub Copilot
Principle · MoneyTreating tools as the full context is a mistake. A context graph helps you reason across the whole company and must include data, synced data, company skills, and processes — not just tools. Relying on tools alone yields an incomplete graph that cannot reason across the company.
AI Engineer · Why Your Company Needs a Context Graph (and How to Build It) — Gil Feig, Merge
Creator perspectiveAlphaFold 2's design was not accidental: it baked in scientific intuition from biophysics and biochemistry — that amino acid residues influence each other rather than acting independently. Baking inductive bias from scientific knowledge into a model makes it much more data efficient because it doesn't have to replicate everything from data. Sal adds that with smaller amounts of data, having more inductive bias in the model helps — you actually need it to get results at smaller scales of data.
explicit
Latent.Space · Why AlphaFold Didn't Solve Protein Folding — Pushmeet Kohli, Google DeepMind & Sal Candido, Biohub
Principle · MoneyEvals are not built once — they're a flywheel that evolves as your customer evolves. As customers move through the journey, what they try to do with your product changes and gets more complex, so evals must stay up to date and reflect exactly what the customer is trying to do right now. Evals must reflect what the customer cares about, not what you think they care about, so get customer input and make sure evals reflect their problem.
AI Engineer · Why 80% Reliability Isn't Good Enough — Felipe Blanes, Amazon AGI Lab
Principle · MoneyThe solution to the benchmark illusion is not better benchmarks but closing the customer loop: get feedback from your customer and feed it back in. Success in the eval flywheel should be defined as what the customer thinks is success, not what you think is success.
AI Engineer · Why 80% Reliability Isn't Good Enough — Felipe Blanes, Amazon AGI Lab
Principle · MoneyIn a world where anything is possible, deciding what to build is more important than ever. Agents cannot replace the human work of deciding what matters, understanding what is being built, interpreting ambiguity, owning the system for the long haul, exercising judgment and restraint at the architecture level, and knowing what not to build.
AI Engineer · Generation Is Cheap, Review Is Expensive: How to Stop Shipping AI Slop — Gabriel Martinez, G2i