A traceable decision trail increases human confidence
You don't need to show all the logs and math behind a single number, but having a trace to follow back a decision increases confidence for humans quite a bit.
Explore
417 results · Most relevant
Results
WHY · 415 of 33339 results · technical analysis
Show the receipts: establish where information came from and why it's correct
When building a harness, show the receipts — where something came from and why it is correct — establishing a causality relationship; this builds trust in the agent's output.
Build feedback loops and gates without relegating accountability
Build more feedback loops so you don't have blind spots, but without relegating any accountability. Build gates for validating things, staying accountable, and correcting course safely.
Align definitions with the model before working
Your definitions of certain things will not exactly match the model's unless you have models trained for you. Go back and forth with your model to align definitions, because what the model knows is not necessarily what you think.
Handing over the keys to the kingdom to something you don't trust is a mistake
Handing over entirely the keys to the kingdom to something you don't necessarily trust to have your best interest in mind is a mistake.
Make harness components work together, not on hopes and magic glue
Make sure all components of your harness work together as they should rather than based on hopes, dreams, and magic glue from an agent orchestrating things.
Figure out what you care about and give each step intent
Figure out what you actually care about and what the functional requirements are for a functioning harness. Make sure the objectives of each step have an intent.
Prioritize trust-building mechanisms over looking correct
Prioritize trust-building mechanisms over just giving something that looks correct when building a harness.
Build the feedback mechanism on purpose so input changes trigger consequential decision-making
Build the feedback mechanism on purpose and have some judgment over making sure there is some consequential decision-making on the fact that an input may change your whole analysis.
Build trust by surfacing implementation information
Build trust with your agent by making sure the information you care about about its implementation is in front of you. The intent of the UI was aggregation of everything: all steps showing where results come from, that it's running SQL, that it's tested.
Gauge user involvement like Claude Code; let agents handle some financial decisions via scenarios
Claude Code has a feature that allows you to gauge how much you want to be involved in the process, and financial harnesses can guide users in the same way. With money, some things can be left for the agent to figure out — for example, if there are a few possible values for something, ask it to do three scenarios. When the agent generates multiple scenarios, review them side by side before making any decision. You wouldn't hand over entirely the keys to the kingdom to something you don't necessarily trust is going to have your best interest in mind.
Escalation path: try skills first, then build a harness for repeated workflows
When trying to use something and not getting the right results, try adding a couple of skills, commands, or extra things first. If you still aren't getting the results or experience you want — finding yourself copy-pasting text from one chat to another, opening new sessions, and it feels convoluted — build a harness. For all the repeated things you had to do three to five times, you design a harness around them once.
With a few possible values, run three scenarios and review side by side
Felippe recommends that when you have a few possible values for something, you ask the agent to do three scenarios and review them side by side before making any decision.
Add external points of reference as notes in the decision record
For verification, add notes in the decision record that bring in an external point of reference for some of the things the model thought.
endorsed advice
Click an underlined changed figure to trace where it came from
The hosts explain that if you want to make a change and click on the underlined figure that changed, it should show you all the way back where this came from and why it matters for the rest of your calculation.
Correct inputs mid-analysis: hover the chip, state the correction, let the harness replan
Francesco demonstrates correcting an input: hover into the chip that guessed how many guests we have, add it to the chat, and say something like 'actually it's not this number, we will get no fewer than 150 guests'. The harness will then rethink the plan, see the consequences of that input into other stuff, and suggest either a modification of your plan or just change the input.
Start in the chat window by describing your financial problem
Francesco directs users to start with the chat window on the right-hand side of the UI to describe the financial problem they want to analyze.
A 'build yours' file in the repo bakes the principles in order to guide building a similar harness
There's a little extra file in the repo called 'build yours' that has all the principles baked in, in order, and will guide you on building a similar harness — prioritizing trust-building mechanisms over just giving you something that looks correct.
Start from the goal and describe your problem to the agent
Start with a chat: describe the problem and what you want to achieve. The harness is designed not to go into yes-man mode but to drill down into what needs to be done.
Combine public data with the user's private data for professional data products
Anekha Sokhal recommends that when building a data product for professionals, you combine public data sources with the user's private data — their trusted analysts, subscriptions, and internal datasets — because personalization is what makes the output relevant. She describes taking these data points together and doing the data cleaning and transformation.
Endorsed by the speaker; no quantitative evidence given.
Protocols you can run
WHY · 0 of 59 published protocols
No published protocol matches this selection.
Source analyses
WHY · 2 of 649 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.