Make the AI Agent Show Its Work
Two companies described showing the evidence behind an AI agent's answer. What Bloomberg's Shawn Edwards and LinkedIn's Prashanthi Padmanabhan said, and three review questions.
Show the user the evidence behind each answer. At an online enterprise AI summit on November 12, 2026, Bloomberg's Shawn Edwards and LinkedIn's Prashanthi Padmanabhan both described it: Bloomberg shows the source, the query and the API calls, and LinkedIn's hiring assistant says why each candidate fits.
Bloomberg's Shawn Edwards and LinkedIn's Prashanthi Padmanabhan both talked about showing users the evidence behind an AI answer, at an online enterprise AI summit on November 12, 2026.
What Bloomberg shows the user
Edwards, Bloomberg's Chief Technology Officer, said one of their principles is that they never want the model to generate a fact or a number. Answers are grounded in more deterministic systems and their sources of truth. The user sees the paragraph or two from the source, the query and the API calls. He added that this mindset is needed for finance and maybe not for other uses of AI.
What LinkedIn shows the user
Padmanabhan, VP of Engineering for Talent Solutions at LinkedIn, said customers are not going to trust AI agents by default. So the hiring assistant says, for every candidate it recommends, why that person fits. LinkedIn calls it "show your work."
Three places BNEDai looks in an AI agent's output
BNEDai's review questions come from the October 7, 2026 session with Women Applying AI. Jacqueline V. Twillie showed three places to look when you want to know what an AI agent did.
- 01
What did it do? The AI agent wrote out what it was doing as it worked.
- 02
Where did it pull from? She said she could read the source the AI agent pulled from.
- 03
What did it write for itself? The instructions and skills the AI agent wrote can be opened and read.
She also showed that the reasoning can be opened after a run and reviewed. The recording does not show a case where the AI agent was wrong. The walk-through is in can I see how my AI agent reached its answer.
What a manager can ask for this week
BNEDai suggests asking the team that runs an AI agent for the source behind a figure the AI agent produced, the inputs it used for one real run, and the instruction it followed, in the words the team gave it. A team that can show all three by pointing at a screen has something behind the output. Ask for them in a meeting where the person who built the AI agent can open each one on screen. If the team cannot show them, a person decides what goes out, as in BNEDai's default for every AI agent. The page on what an executive sponsor should ask after a first AI agent test works through the questions in more depth.
Frequently asked questions
What does Bloomberg never want the model to do?
According to Shawn Edwards, generate a fact or a number. Answers are grounded in more deterministic systems and their sources of truth.
What does LinkedIn's hiring assistant do?
For every candidate it recommends, it says why that person fits. LinkedIn calls it "show your work," as Prashanthi Padmanabhan described.
Where did the three review questions come from?
From Jacqueline V. Twillie's October 7, 2026 session with Women Applying AI.
Source: remarks by Shawn Edwards, Chief Technology Officer, Bloomberg, and Prashanthi Padmanabhan, VP of Engineering for Talent Solutions, LinkedIn, at an online enterprise AI summit, November 12, 2026. The three review questions and the weekly ask are BNEDai's.
Reviewed on November 20, 2026.