35,800 results · Most relevant
The speaker advises that if cost is a concern, build reusable agents and have people solve problems the same way rather than reinventing the wheel, to drive down your overall bill. Reusable components drive up cache rate because you get exact matches between turns on common data.
Mansour advises that when building features for humans, teams should also expose them as skills and tools for agents, based on Atlassian's experience converting human features into agent-usable tools.
Pedro Lopez advises that when implementing OAuth for MCP, keep offline long sessions stored separately from the web app with their own set of life cycles, since users expect long-lived sessions through MCP servers — more like a mobile app where they don't have to think about authenticating.
Pedro Lopez warns against putting too much content into a skill, which bogs down the agent's context. Instead, structure skills with a main skill pointing to reference files in a skill directory, cutting context load and leveraging agents' progressive discovery.
Pedro Lopez advises that consistency is critical in agent-facing CLIs — if the CLI uses noun-then-verb structure, every resource should follow it. If agents repeatedly call the wrong command, that usually means something isn't obvious enough in the contract; consider making that behavior the contract.
Don't lock into a 'modeler' or 'data generation person' identity — that mindset can cause failure. Understand the problem you're trying to solve, then be flexible in the solution space: do the modeling if modeling is needed, collect data if data is needed. The broken pattern in ML research was optimizing models while ignoring whether the data was adequate.
Modelers default to whatever data is available, and data generators default to what is easy to generate — both paths can lead somewhere negative. Instead, define the problem first and identify the data actually needed to solve it; this requires model builders and data generators working together openly.
Automattic is beginning a guides program (like other companies' champions programs) where identified top practitioners help spread knowledge. Program graduates also run local AI get-togethers (small regional events and workshops, about 10 done so far), and the company organizes AI ride-alongs: small workshops showing how someone in a similar role uses a tool, since seeing a peer use it is easier to grasp than discussing concepts.
Patel advises founders to let learning drive their decisions rather than ego. Grind it out when the business model actually makes sense, but don't grind it out in a business which doesn't fit your ambitions — grinding in a business that doesn't fit your ambitions is a mistake. If you want a good lifestyle business that provides well for your family, keep doing that — it's a great answer. Check whether your ambition aligns with what you're actually doing right now — whether you'll feel fulfilled looking back at 80.
opinion
Will Lyon recommends using the context graph to create grounded, executable skills — the next step after memory.
Abhi Arya recommends adding instrumentation — logging prompts, exported sessions, override points — and flowing that data back into the agent's long-term guidance so the system improves by building more pipelines without hand rewriting. If you hand-rewrite agent guidance instead, you lose the self-improving loop per org or per user.
You can provide an annotation in Figma by pressing the Y key to add additional context; in design mode an annotation can be a category of interaction, e.g. 'when scrolling to this section, perform the embedded animation', and when you pass off a frame to Figma Make or an LLM, the interaction annotation gets packaged along with it.
Create additional keyframes in the prototype so it animates to each respective state when played.
Mickey demonstrates exporting a transparent animation: hide the fill of the frame, export as WebM at next size (current dimensions), high quality, 30 frames per second. He notes WebM honors transparency unlike some other formats, and shows importing the exported WebM back into Figma where it plays as a transparent video. He adds WebM is especially useful if building for web and wanting to try transparent video.
Mickey demonstrates simulating a set number of loops: extend the animation to a longer duration (e.g., a two-second animation taken to six seconds), then right-click the animating property and choose loop track. Loop track proliferates the keyframes as ghost keyframes so the animation continues looping across the extended duration; right-click each animating property so multiple properties loop together.
Matt Owens advises agency leaders to keep the business driven by good work and creativity — putting craft, design, story, and strategy at the center, with finances and operations in service of that, not the reverse. He warns against the drift where work becomes a fulfillment of a financial mandate: there is always tension between doing great work and the business side, and letting the mandate win kills the agency's purpose.
Matt Owens argues that starting your own business should not be mysterious — if a guy can start a record label, a guy can start an agency. It takes the right level of energy and not overthinking it: commit ('I'm committing to this. I'm going to do this') and let it come out through the effort. He also notes the startup culture he worked in had an absence of analysis paralysis — people were figuring it out.
Matt Owens says to treat AI in brand systems like sriracha: a nice additive ingredient integrated into the whole recipe, not a replacement for anything. It's not 'the ozempic of brand building'. You still need a human After Effects person to make a sizzle reel that's face-melting and compelling — AI-generated assets can fill in due to time constraints, but AI is an ingredient, not a replacement. Use AI to get mechanically to a creative output faster, but the crafting of a beautiful, resonant human thing remains your responsibility.
Matt Owens urges creative professionals to always put themselves in a position of tension and challenge and not get too soft — getting through the other side gives you the spark you had when you were 19. Whatever that inner fire is — the need for challenge, struggle, and problem-solving — keep it around as much as possible. Feeling the burn is the reason you get into creative work, period; the pain and struggle is the secret sauce.
Matt Owens recommends building technical skills ahead of the mainstream — having capabilities others don't yet have gives you a competitive advantage. He describes how timing his graduation from Cranbrook (1995) coincided with the internet becoming commercially viable, giving him an advantage: he already had built his own website and had an email address when others didn't.
Choose sources → to get a library of your own and a daily brief.