Interesting Topics

November 10, 2026

AI · Measure the whole task before funding a rollout

In METR’s randomized study, 16 experienced developers took 19% longer to finish tasks in familiar repositories when they could use early-2025 AI tools. Before renewing AI licenses for sourcing analysts, compare time to an approved recommendation on similar cases with and without the tool. Include fact-checking and rework in the time measured. Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity — METR, July 10, 2025 · Foundational

November 9, 2026

AI · Measure who benefits before expanding a pilot

In a study of 5,179 customer support agents, access to a generative AI assistant increased issues resolved per hour by 14% on average. Gains were greatest for newer and less skilled agents. Before expanding an AI pilot for supplier inquiries, compare resolution rates and reopened cases by buyer experience. Use those results to decide which teams should get the tool first. Generative AI at Work — NBER, April 2023, revised November 2023 · Foundational

November 8, 2026

Supply chain · Trace dependencies beyond the ship-from country

The OECD’s input-output analysis found that China accounted for about 37–47% of global value added in five manufacturing industries in 2024. Before approving a second source for a critical component, trace its key inputs through upstream producers and test whether a disruption in one country could interrupt both suppliers. Economic security in global production: Dependencies and vulnerabilities in GVCs — OECD, November 6, 2026

November 7, 2026

Supply chain · Use global pressure to trigger a local check

The New York Fed staff paper found that its global supply chain pressure index tracked producer price inflation in the United States and euro area. When the index rises, check actual lead times and freight quotes on exposed lanes before committing cash to extra inventory. The GSCPI: A New Barometer of Global Supply Chain Pressures — Federal Reserve Bank of New York, May 2022 · Foundational

AI · Test when an agent should stop

Anthropic found that among the longest Claude Code sessions, autonomous work before a pause rose from under 25 minutes to over 45 minutes in three months. Before giving a sourcing agent a longer run, test whether it stops when required evidence is missing or an action needs approval. Measuring AI agent autonomy in practice — Anthropic, February 18, 2026 · Foundational

November 6, 2026

Procurement · Find out why only one supplier bid

The OECD recommends reading single-bid rates alongside market concentration and barriers to entry. In your next category review, sample single-bid awards and record whether the supplier market was thin or the tender discouraged viable bidders. Use that distinction to decide whether to qualify new suppliers or change the next bid package. Competition in public procurement — OECD, October 26, 2026

AI · Test whether an agent can pick up where it left off

In Anthropic’s coding experiments, a progress log and incremental work gave each new session a record of what had been done. Before expanding an AI pilot to multi-day supplier research, interrupt a test run and check that it can resume from its evidence and open questions without repeating work or treating unchecked claims as verified. Effective harnesses for long-running agents — Anthropic, November 26, 2025 · Foundational

November 5, 2026 · Three foundational reads

Supply chain · Where diversification breaks down

The IEA distinguishes mining concentration from refining concentration. Two suppliers can look independent at the contract level while relying on the same upstream processing geography. For a critical input, map the refiner and country behind each qualified supplier before treating the second source as independent. IEA, Global Critical Minerals Outlook 2026, executive summary

Procurement · Measure what the sourcing process delivers

The OECD separates efficiency, compliance, and strategic outcomes in its public procurement scorecard. For an enterprise category review, borrow that separation. Track cycle time, control adherence, and realized business value as separate measures so a fast award cannot hide a weak outcome. OECD, Public Procurement Performance

AI · Match oversight to the decision

NIST’s generative AI profile describes how use context affects governance and oversight. Apply that distinction before putting AI into procurement workflows. Test an assistant that summarizes bids differently from one that recommends awards, and specify who can approve an action and how the decision will be recorded. NIST, Generative AI Profile