35,595 results · Most relevant
Gifted kids have the most working memory slots but do everything in their head and never write down their work; every gifted kid eventually hits a wall because some math can't be done in your head, and they learned bad habits instead of good ones. The remedy: write down your work on multi-step problems.
Kent argues that with AI agents, every engineer is now a team lead over however many agents they can run at once. Just as a team lead's traditional job was to make the system efficient for engineers — good testing, good architecture, obvious Lego blocks — it is now the engineer's job to make sure the playground is nice so agents don't make a ton of mistakes. Engineers should take great care in the system they create for their agents to ensure success.
The Instagram chatbot security failure could have been prevented at both the harness layer (just-in-time) and the eval layer (ahead-of-time); ideally evals would have caught it before deployment. An eval can verify that a chatbot confirms the user is who they say they are, regardless of how much they ask to change or where they log in from.
The speaker advises the audience not to index on one eval or one test — it should just be one signal of readiness to the platform — and endorses the principle that no eval or test is a measure of someone's complete abilities, which should always be clear going into it.
Elena Verna argues that if you don't change your organizational design to take advantage of AI skills and the people using those tools, you're not taking advantage of the biggest change happening in the system right now.
Elena Verna recommends decoupling pay and status from headcount when organizing your team, calling it important.
Jeff Smith recommends piping existing compiled reports into the database or MCP server and presenting them as tools or APIs the agent can call. Describe each report with sample questions, meaning, and all parameterized inputs so the agent can pick the right report. Hard-coding the code underneath reports means the LLM might pick the wrong report but won't generate the wrong SQL. Expose a List Reports tool instead of one tool per report, because 500 tools would blow out the agent's context. Use a key phrase to ask the agent for any report that looks like a description, get a report list back, then execute one of those reports.
hedged
Jeff Smith recommends using a test environment that mirrors production so agents can investigate performance issues safely. He also suggests using Oracle DataGuard's read-only standby instance to offload reports and read-only queries, where performance issues likely replicate, and letting the agent interrogate that safer environment to investigate and try out different performance tricks.
Jeff Smith recommends writing three or four multi-purpose tools that cross categories with options, instead of 15 single-use tools. He describes writing a custom tool with six different actions the agent is allowed to do on his behalf, including collecting statistics that profile the data and generating an AWR report.
Jeff Smith recommends annotating tables and columns in the database, especially columns that jump between units of measure and time, and columns that don't contain what you think they do. Properly annotating tables makes it easier to one-shot SQL queries when doing natural language-to-SQL exercises. The nice thing about the schema annotation feature is it's all in plain English, it's in the database, and it's going to help the humans too.
Jeff Smith recommends that when the agent needs to learn the data model, you show it the SQL behind existing reports as a blueprint for how the business and data model work. With proper OAuth 2 scope or role, the tool can show the agent the SQL code behind a report so it learns how the data is organized and one-shots the right query.
Jeff Smith recommends using a schema info tool that returns a compressed dictionary of attributes in JSON, instead of the agent running queries like 'show me all the tables, show me all the columns, show me all the comments' and reading five rows out of each table. The tool also has a filtering mode where you can give it keywords so it looks for specific things in the data.
Bakken advises that you need to do easy on the easy runs; otherwise you will start to accumulate one session into the other, undermining the threshold load the week is meant to deliver.
Bakken endorses a balanced training approach as the way to both be in better health and run faster.
The host says the biggest lesson from the book is that if you can keep your muscles healthy and feeling springy, it is highly likely you can continue to get miles in and be less prone to injury. Bakken describes the system across the week: you're trying not to accumulate tone throughout the week but go up on the hard days and down on the easy days.
Consequential creator advice requires independent verification before Today can promote it.
Dr. Marius Bakken advises keeping muscles healthy and feeling springy, which makes it highly likely you can continue getting miles in and be less prone to injury. The host echoes this as the biggest lesson from the book.
Consequential creator advice requires independent verification before Today can promote it.
Bakken endorses that keeping muscles healthy and feeling springy makes a runner highly likely to be less prone to injury.
Bakken recommends finding a way to do the cardiovascular work without constantly going into the race-pace zone, i.e., using sub-threshold work as the backbone of training.
Bakken recommends doing training within a strict framework instead of doing it by feel to unlock performance gains.
When your champion asks for a contract, walk them through the next steps live: confirm when they will submit it to legal, when legal will return first-round redlines, and draft the follow-up email on the spot while sharing your screen. Pre-schedule the kickoff call with the implementation team before the contract is signed, noting that a signed contract is required to hold it — this makes the champion hold legal and procurement accountable to the agreed timeline.
Choose sources → to get a library of your own and a daily brief.