RecommendationThe agent's rules are configured by change type in the PR and know what layer one already checks, so the agent focuses on the risk those checks cannot catch: query anti-patterns, sensitive data asks, and consumers outside lineage. For schema changes it walks pipeline definitions and access grants in the repository to find unregistered consumers — a tedious search a person would otherwise do by hand.
AI Engineer · 60% of Our Data PRs Now Merge With No Human Review - Chetas Joshi, Robinhood
Recommendation · MoneyJack Wotherspoon recommends tuning managed agents to the workflow by setting the model via the one-line parameter in the agent config: use Gemini Flash when the workflow needs to be fast and efficient, and a frontier model like Gemini 4 Argon for deep reasoning or heavy analysis. He notes Gemini 4 Argon was released yesterday and is especially good at financial tasks and data analysis across a wide range of documents, and recommends testing it out.
hedged
AI Engineer · Build Your Own Team of Managed Agents - Jack Wotherspoon, Google DeepMind
Recommendation · MoneyJack Wotherspoon recommends using webhooks with triggers so a managed agent can send a Slack message, for example querying the team's JIRA board every morning and sending what everyone is working on.
AI Engineer · Build Your Own Team of Managed Agents - Jack Wotherspoon, Google DeepMind
Recommendation · MoneySam recommends using semantic caching to serve similar prompts from a built-in Redis store instead of sending each request to the LLM, achieving zero cost and zero latency for cached responses, plus a nicer user experience. Configure the similarity threshold on a 0-1 scale and flex it per agent, per audience, or per user.
AI Engineer · AI Agents Don't Read Your Policy Docs. They Hit Your APIs — Gravitee
Recommendation · MoneySam recommends controlling token spend and costs through the gateway — applying rate limits (e.g., three requests per minute) so excess requests are blocked before reaching the LLM and spending tokens, rather than being blocked downstream where tokens are already spent.
AI Engineer · AI Agents Don't Read Your Policy Docs. They Hit Your APIs — Gravitee
Recommendation · MoneySam recommends exporting gateway telemetry to any downstream tool such as Datadog, Splunk, or Prometheus (if running in Kubernetes) rather than being locked into a black-box platform, and verifying the platform provides full logging and transparent telemetry.
AI Engineer · AI Agents Don't Read Your Policy Docs. They Hit Your APIs — Gravitee
Recommendation · MoneyThe way Warp does spec-driven development — which Lloyd recommends — is having an agent write a product spec and a tech spec: the product spec describes product invariants you're building towards, and the tech spec describes the architecture and the shape of the code.
AI Engineer · Software Engineering Is Becoming Factory Engineering — Zach Lloyd, Warp
Recommendation · MoneyFor certain types of apps, Lloyd says the verification step is very important: use computer use so the computer actually uses the code the agent produced and produces videos and screenshots. And still use CI/CD, obviously.
AI Engineer · Software Engineering Is Becoming Factory Engineering — Zach Lloyd, Warp
Recommendation · MoneyLloyd recommends having an agent do code review first, and then over time treating human code review as a risk management exercise of deciding when to bring in humans — but you want to keep a step in the factory where humans can do code review.
AI Engineer · Software Engineering Is Becoming Factory Engineering — Zach Lloyd, Warp
Recommendation · MoneyLloyd recommends treating triage as a really important step: you want an agent looking at issues as they come in and, if an issue is easy and unambiguous, just implementing it — this is how you can actually get going with a factory. If an issue is hard, he recommends having an agent that produces specs. In the cloud software factory loop, ideas come in at the top, agents do triage, and if something is complicated they write a spec.
AI Engineer · Software Engineering Is Becoming Factory Engineering — Zach Lloyd, Warp
Recommendation · MoneyJohn Stecher says lean into modern software development harnesses because they let you go from concept to a workable proof of concept remarkably quickly; use tools like Fable to define a concept collaboratively with the product person or stakeholder before writing code. He encourages all his engineers to frequently just get to POC rather than building PowerPoint decks, and says don't worry about PowerPoint or anything else.
AI Engineer · Why Waiting for the Next AI Model Can Pay Off - John Stecher, Blackstone & Scott Worland, Norm AI
Recommendation · MoneyJohn Stecher recommends using agents to write functional and performance test cases and version forward libraries in the software development lifecycle, and reports that having an agent renovate code and source code has made all the difference in the velocity his team delivers at.
AI Engineer · Why Waiting for the Next AI Model Can Pay Off - John Stecher, Blackstone & Scott Worland, Norm AI
Recommendation · BodyIf you only have two days a week for cold exposure, do short 1-2 minute dips, get out, and get back in for another couple of minutes, working toward the 11-minute weekly total; the temperature changes themselves strengthen cells.
hedged
Andrew Huberman · Essentials: Cold & Heat Exposure for Metabolic Health | Dr. Susanna Søberg
Recommendation · BodyDr. Søberg endorses that putting just a hand in cold water can activate brown fat through a temperature change on the skin alone. Going outside in a t-shirt or wearing cooling vests for 10 days will grow brown fat, and you don't have to start with a cold shower or cold plunge — just exposing yourself to cold wind has also been shown to activate brown fat.
Andrew Huberman · Essentials: Cold & Heat Exposure for Metabolic Health | Dr. Susanna Søberg
Recommendation · MoneyIf a specific part of the workflow consistently does tool calls to get information, the eval has to be designed for tool calling and not for a different process. Designing an eval for a different process than the workflow element being tested means not knowing the tool call ability and accuracy at that point.
AI Engineer · If Your AI Scores 100%, Your Eval Is Wrong - Ramakrishnan Lokanathan, eslams.com
Recommendation · MoneyJeff Smith advises keeping the code underneath a report more or less hard-coded so the LLM might pick the wrong report but won't generate the wrong SQL. This removes one place the LLM can mess up.
hedged
AI Engineer · Your AI Agent Wants Your Data. Here's How to Lock It Down - Jeff Smith, Oracle
Recommendation · MoneyJeff Smith advises using tools that consume Confluence pages to generate context and document schema metadata over top of the metadata that's already in your database, and delivering to the LLM the information about the database environment you're talking to, instead of relying on it to generate queries to reverse engineer and hack your database to look for data.
AI Engineer · Your AI Agent Wants Your Data. Here's How to Lock It Down - Jeff Smith, Oracle
Recommendation · MoneyJeff Smith advises putting time into describing each report you expose to the agent: describe what the report means and all the inputs, since reports are parameterized, and include sample questions so the agent knows which report is good to run for a given question.
AI Engineer · Your AI Agent Wants Your Data. Here's How to Lock It Down - Jeff Smith, Oracle
Recommendation · MoneyJeff Smith advises exposing just two MCP tools for reports: List Reports and Execute Reports. Do not expose 500 reports as 500 tools — that would blow out the context window and make the tool list useless to the agent. Use a key phrase so the agent can request any report matching a description, get a report list back, and then execute one of those reports to get the data.
AI Engineer · Your AI Agent Wants Your Data. Here's How to Lock It Down - Jeff Smith, Oracle
Recommendation · BodyBakken endorses that for a 5K, the X session pushes the intensity above the anaerobic threshold, going closer to race pace or even above it to get a neuromuscular effect, while avoiding triggering anaerobic domination of the training week.
Garage Strength · The Genius Who Solved Endurance Training w/ Dr. Marius Bakken