Set up the workshop materials from the QR code or repo
Francesco instructs attendees to scan the QR code or go to his repo and follow the instructions to set up the workshop materials.
Explore
417 results · Most relevant
Results
WHY · 415 of 34614 results · technical analysis
Verify the setup, then run the second command to open the UI
Francesco explains that if the setup works you should see the UI opening and success in the terminal, and that the second command opens a well-designed UI to see how the harness behaves in the real world.
Open a new branch conversation from a step for asides
Francesco demonstrates that you can click on step five to open up a new branch conversation.
Switch use cases by deleting the database and reopening the UI
Felippe explains that to switch between the wedding example and your own use case you delete the database in the my and var folder, then reopen the UI with the previous command.
Distinguishing buyers from non-buyers among people with the problem
Anekha Sokhal draws a distinction between people who have the problem but don't want to buy and people who have the problem and want to buy. This frames how she thinks about finding clients.
Harness value: more control combining deterministic bits with non-determinism
The value you get from building the harness is more control, with the deterministic bits together with the non-determinism. By 'determinism,' Felippe means the back-end code — the Python, whatever is being used — that the harness provides.
Golden examples are ground truth, but carry selection bias and no domain is isolated
Golden examples give you your ground truth for making sure that you are within the right space, but no domain is isolated. Golden examples given to an agent in a harness are based on biases and selection bias from the domain, picked for the domain and thought to be good for the domain.
Accountability mechanisms in harnesses reduce second-guessing stress
The hosts describe that having accountability mechanisms in harnesses reduces a lot the amount of stress of going in your head wondering 'do you think it made the wrong assumption here'.
There's no single thing that earns user trust; different users value different evidence
There won't be a single thing that works to earn the user's trust at any point — for some people verification is more important, for others seeing actual calculations and numbers matters more; the process guides and covers a multitude of possibilities.
endorsed
Your definitions won't match the model's unless it is trained for you
Francesco notes that your definition of certain things will not exactly match the model's definition unless you have models trained for you.
Approving every calculation or reviewing every PR line by line is toil
Francesco describes the opposite extreme: approving every calculation or reviewing every PR line by line that your agents ship doesn't feel great, leads to a lot of toil and unnecessary work, and isn't really well-written work.
Wedding planning demo prompt with payments, fixed costs, and unknown guest count
Francesco demonstrates a realistic prompt: he has a wedding and needs to make three payments — one for $6,000, one of unknown amount, and one for the rest — with some fixed costs like a DJ and decorations, but he doesn't know how many people will attend, so he doesn't know the per-seat dinner cost or how to make sure he can pay in time.
Take what you build home and review it
Take everything you build today home and review what you did, what you liked, and what you didn't.
Expect things to break and call the presenters over
Expect something to break during the workshop and call the presenters over so they can fix it with you. Raise your hand if something is not working or you are confused, and the presenters will float around the room.
Prioritize client meetings over hiring, and scale conversations with webinars
Anekha Sokhal recommends prioritizing meeting clients over hiring when you have continuous inbound from LinkedIn, since hiring can be passive but client meetings require active pursuit. When there is too much inbound to handle one-on-one, she suggests hosting a webinar to scale the conversation instead of only doing individual meetings. She also describes the bank practice of round-table meetings — a CFO meeting five funds at once — as a format worth replicating for conference networking.
Practical advice from the speaker's current operating experience; the 5-on-1 format is her recollection of bank practice.
AI engineers building systems alone and adoption struggles
Anekha describes AI engineers needing to build a system that would typically be done by 10 people, asking whether they need a swarm of agents and what happens if something breaks. She also describes AI engineers who built a system people said they needed, it's in their email, but no one is clicking it while they still complain about the problem.
Distinguishing people who have the problem and want to buy
There is a distinction between people who have the problem but don't want to buy and people who have the problem and want to buy. The speaker has not found any person who successfully matches people who have the problem and want to buy.
AI engineers are overwhelmed: 'you are the team'
Anekha Sokhal describes AI engineers feeling overwhelmed by management pressure because 'you are the team' — they are expected to build systems alone that would typically be done by 10 people. They worry about whether they need a swarm of agents and what happens if something breaks. A common failure mode she cites: someone builds a system people said they needed, it sits in their email, and no one clicks it — even though they still complain about the problem.
Anecdotal reports from people the speaker has talked to; the '10 people' comparison is her framing.
Recommendation systems are underappreciated by engineers
The host notes that as a curator of AIEs he has an LLM Rexys track every year, and engineers often don't appreciate how important recommendation systems are — the track is not well attended even though the same people power Rexys for YouTube, Netflix, Instagram, and TikTok. Anekha agrees that Rexys is one of the most important parts of AI.
Engineers undervalue recommendation systems despite their outsized importance
Anekha Sokhal claims that engineers often don't appreciate how important recommendation systems are: they power YouTube, Netflix, Instagram, and TikTok, yet conference tracks on them are poorly attended. The host adds that recommendation systems seem easy but are arguably the most important system in an entire company.
Observational claim about conference attendance and industry attitudes; the 'most important system' framing is the host's opinion.
Protocols you can run
WHY · 0 of 60 published protocols
No published protocol matches this selection.
Source analyses
WHY · 2 of 670 current source analyses
People worth following for this
The creators behind most of what cleared.
WHY · 2 creators you don't follow account for 415 cleared insights
- AI Engineer
344 cleared insights match
- Latent Space
71 cleared insights match
Choose sources → to get a library of your own and a daily brief.