Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode!

Five evaluators read the same AI-generated summary before scoring a tender. By the time the panel meets, their scores are remarkably similar. Did five people independently reach the same decision, or did one AI interpretation shape five opinions?

In October 2026, Anthropic published a study of 50 professional analysts working with an AI agent that had been secretly given a randomly chosen preference. Across four analytical choices, the analysts followed the agent's preferred option 86% of the time, compared with around 50% if it had no influence. This episode looks at what that finding does and doesn't tell us, why a questionable assumption is harder to spot than a factual error, and what procurement's practice of individual scoring before consensus can teach any function using AI for important decisions.

In this episode:

• Why approving an AI recommendation isn't the same as deciding

• What the Anthropic study found, and its limits

• Why independent first views give an evaluation panel its value

• A hypothetical sourcing decision where a $120,000 price saving becomes around $44,000 in extra annual cost, before the one-off qualification cost and any resilience benefit

• What a 2026 study of trainee teachers suggests about forming your own view before seeing AI advice

• Why this isn't an argument against AI, and how to test when AI improves your decisions

View Before Review, for decisions that need judgement:

1. Before reading the AI's recommendation, record a short initial view: what you would choose, your main assumption, and what evidence would change your mind.

2. Ask the AI for its analysis, the assumptions behind its recommendation, and the strongest reasonable case for a different option.

3. Compare the positions. Which differences are facts you can check, and which are assumptions that need a decision-maker's judgement?

4. Record the final choice, why you made it, and how you will review the outcome.

The tender panel and manufacturer examples in this episode are hypothetical.

Sources

Anthropic, How AI agents shape experts' judgment calls (27 October 2026):

https://www.anthropic.com/research/agents-judgment-calls

Yan, D., Independent judgment before viewing generative AI recommendations reduces susceptibility to misleading advice: a randomized scenario-based experiment with preservice teachers, Frontiers in Psychology (28 September 2026):

https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2026.1936522/full

Procurement Services SA, Evaluation Process Guideline, version 1.1 (2023):

https://www.procurement.sa.gov.au/documents2/guidelines2/Evaluation-Process-Guideline.pdf

Episodes and resources: https://www.kaizina.com.au/podcast

Procurement at AI Speed is hosted by Zina Amir Moafi and produced by KaiZina.

General information only, not legal or financial advice.