Test fixes against a dataset built from the spans that caused the failure initially
When fixing an issue, Duggal recommends testing the fix against a dataset built from the spans that caused the failure initially.
44,661 results · Most relevant
When fixing an issue, Duggal recommends testing the fix against a dataset built from the spans that caused the failure initially.
Prompt the agent with a skill to pull down traces from a specific agent, find issues from evals and annotations, do its own analysis, and suggest future evals to prevent recurring issues. Then instruct it to build the recommended evals and run experiments against the spans that caused the initial failures, using the Arize dataset, evaluator, and experiment skills.
Set up the Arize CLI by copying a single line of install code, create a profile connecting it to where your data is stored, then set up skills in your coding agent (method varies by agent, e.g., Marketplace in Claude Code). Get your code instrumented so all agent traces flow into the observability platform.
Josh Woodward recommends thinking about planning as almost like a rolling six months — a short, continuously updated horizon suited to fast-moving AI work.
Get outside the building and show people prototypes. When talking to someone about a product, you know if they're having a good time, if they're into it, or if they're just bored out of their mind. A way someone else described the 'eyes light up' signal: their pupils dilate.
When an agent spirals out of control without reaching an answer, an AI gateway lets you point it back at what it was doing and ask what happened and why it didn't work. Cheaper models are usually not great at stopping themselves but are great at analyzing where they went wrong after the fact.
Remy advises matching the AI interface layer to the team: engineering teams likely won't code via Slack threads, but marketing teams might work entirely in Slack asking data questions.
Remy recommends that when a model provider offers OAuth, such as Anthropic OAuth or Bedrock Codex, you configure it on your AI gateway alongside standard API keys.
Remy recommends that when an agent spirals out of control and doesn't reach an answer, you point it back at its own logs and ask it to analyze where it went wrong.
Peter Werry recommends creating a tool that does the query and, as an argument to the tool, accepts a username, so you can fail right there if the username is not correct or malformed.
The guest says the first thing to ask someone starting hybrid training is how much they can actually train. He warns that people overestimate their availability — claiming five conditioning sessions and four gym sessions while barely making it to the gym twice per week. Realistically, if someone says they'll train five times a week, it will be four; if they say four, it will be three. When planning frequency, look at the realistic range and ask what's acceptable within that range.
Pushing top-end clean weights in the off-season is fantastic when it doesn't interfere with in-season training. If you're practicing cleans and making progress while other strength ratios are sorted, you'll get a lot of what you need — e.g., going from 0.8x to 1.1x bodyweight power clean between off-seasons is fantastic progress.
the 0.8x to 1.1x example is flagged as a possible numeric error
Hill sprints let a non-sprinter athlete sprint all-out — open up and hit sixth gear — without tearing their hamstring. Aim for fast, long strides with a little of the front knee coming up. Some track coaches criticize high hills, saying you need a tiny gradient, but the speaker disagrees for non-sprinter athletes.
the track-coach criticism is a quoted position the speaker disagrees with
The speaker recommends pulling pre-built pipeline agents off the shelf from the GitHub Next organization's Agentix project, which has 30 or so different pipeline agents. Some do code deduplication, and others are different types of reviewers, such as a QA reviewer or one that just makes jokes.
Add full text search to the query engine to handle cases where an exact title match returns no results.
To test the benefits of context bundles without wiring up Unblocked, use the open source simulator to create a context bundle on a task-by-task basis and run an A/B test against the same task to measure the benefit of context for your agents.
After learning that whiteboard users typed 'group my sticky notes into common themes', Atlassian let users select cards and trigger the grouping with a button instead of a prompt; automate common prompts into buttons as one of the chat-to-UI patterns.
Atlassian runs AI Builders Week every couple of months, a four-day program across the organization synchronized so everyone takes the same time; juniors come in and share all the new tools and techniques, and seniors come in and talk about quality control, AI slop, and how to build differentiated AI rather than generic AI.
Expose your app's skills via MCP or a CLI so customers' external agents (e.g., coding agents) can use them across their other SaaS tools; agents on your platform need to work with the rest of the customer's world.
Tune the zero-copy retrieval flag and the spread scheduling mechanism flag as part of optimizing the data pipeline.
Choose sources → to get a library of your own and a daily brief.