After an agent drafts work, you must review, adjust, and inspect it
After an agent drafts work, you must review, adjust, and inspect what the agent built to make sure it is bulletproof before it gets sent out.
45,937 results · Most relevant
For understanding everyday Spanish through listening and conversation.
After an agent drafts work, you must review, adjust, and inspect what the agent built to make sure it is bulletproof before it gets sent out.
Shipping is not where the work ends — tax code changes and a lot is changing in the industry in general, so skills need ongoing maintenance.
The agent must understand how your firm delivers advice — it can't just deliver generic recommendations because your firm may be more conservative than what an off-the-shelf agent would deliver.
hedged
The trick to mushroom growing is to overwhelm any mold or bacteria that could potentially still be alive on the substrate. A farmer's job is to make sure no mold or bacteria grows and only the fungus grows. The whole point of pasteurization is to kill competing molds and bacterias.
hedged
The speaker argues there is no oracle for specification correctness other than the user: the LLM lacks implicit domain knowledge, so only the user can guarantee the spec is correct. The user should check every part of the snippet and rulebook before sending it to the LLM, and the pipeline is 'the LLM translates, the solver checks, and the human decides'.
Datasite's build-versus-buy philosophy is to spend heavily with vendors intentionally so the most talented internal people work on things vendors cannot easily handle, not on problems vendors are dedicated to solving every day.
hedged
NNgroup states that using multiple selection indicators improves the visual salience of the selected tab and reduces confusion over which tab is displaying its content.
Nikhil Harithas argues that agent systems should be designed so agents are indifferent to the compute they run on, a recurring theme in agent infrastructure design.
hedged
Nikhil Harithas recommends designing agent infrastructure to manage computers rather than sessions, because tying compute to a session concept creates vendor-specific lock-in that makes customers uncomfortable hosting the software.
hedged
Nikhil Harithas argues that for agentic applications, the compute plane should be separated into its own architectural layer, distinct from the traditional data plane and control plane.
Legora picks the safest possible approach to data privacy and restricts itself as much as possible because there should be no risk at all; they know the 10-20 probable failure modes and classify confidential data into one of those types as a helpful signal, plus classifying into task types and practice areas.
hedged
Dharmesh Shah argues that picking ideas is one of the hardest things, especially for first-time founders, because it is almost impossible to know a priori what the quality of an idea is or its probability of success. The only real way to know if an idea is good is to actually test it, and the thing that ends up succeeding will often resemble the original idea but will be different.
Dharmesh Shah's generalization: the people that win are the ones that have the shortest iteration time and the cheapest learning time. Related: it's a great time to do a startup because AI lets you iterate so much faster, and AI lets you learn lessons that might have taken six months to two years much faster.
hedged
The core problem statement: make culturally rooted products without making costumes. Most attempts at Indian design start looking like costumes — an anti-pattern. The modern Indian has moved on from maximalism, snake motifs, and heavy red/blue detailing; there was no defined design language for the modern Indian. The brand goal: make a design or brand that looks modern, appeals to the contemporary Indian person, and also has a global aesthetic.
The fifth and final element beyond workers is craftsmanship — representing handicraft and the artisans who worked hard on it in a culturally elevated design company. To elevate craftsmanship: understand what it means, look at it from a design perspective, put in thought and effort to contemporize it — that is the only way to put Indian design on the global map.
Molly Graham argues that getting rid of whole management layers inside bigger companies is a huge mistake that will bite people long term, because there are no signs management matters less — if anything it matters more. Management makes people feel seen and supported and helps them figure out whatever mess they're in, and taking care of managers is a reminder that you have a lot of leverage there.
hedged
Molly Graham says managers and leaders more than anyone need to hold on to the definition of accountability and what good means — deciding what is right to give away to AI and what humans still need to own. They will have to role-model this and create standards, in their own behavior, how they lead, and how they talk about it.
Molly Graham argues that people should grieve the loss of work they loved and were extraordinary at, rather than skipping straight to excitement about the future — grief needs a funeral, or whatever ritual lets you be sad. After mourning what was, she says to ask what you could love and be extraordinary at again in the new world where AI does X, Y, and Z.
hedged
Molly Graham argues the future will be defined by people willing to take the stance of learning and throw themselves into change, scary and unknown as it is. When the world changes fast, what you can learn by tomorrow matters way more than what you know today, so prioritize learning speed over accumulated knowledge when evaluating yourself and others. When you notice yourself defending and fighting for what you know during AI change, treat that as an instinct to override, not a strategy to follow. She also advises tossing out what you knew six months ago because something new is true today.
Molly Graham argues that work requiring judgment should not be outsourced to AI, and that each person is phenomenal at a specific set of things they should amplify with AI rather than hand off to what she calls weird robots that behave like summer interns.
Choose sources → to get a library of your own and a daily brief.