--- schema: 1 type: video title: 'Containment Is Lying to You: Measuring AI Agents Like Employees — Druid AI' description: >- Two AI interns. One handles 96% of conversations alone, the other only 80%. The dashboard picks the wrong one. Druid AI's team, including co-founder and chief product officer Dan Bălăceanu, shows why containment rate, the metric on every agent dashboard, can point you to the wrong agent. In their bank example, the intern who escalated more often was handing senior bankers customers who were about to refinance. They walk through a performance review framework built with their customers: seven pillars split into what the agent is configured with and what it does at runtime, with governance across both. Twenty-six components are each scored 0, 1 or 2 (missing, defined, or actually governed and tracked over time) and weighted by business impact, with governance and mandate weighted highest. They cover error tolerance, risk, compliance, observability, drift, hallucination, tone and sentiment, the agent's mandate, scope, business case and budget, a shared platform, and validating outcomes. They also show a dashboard that stays green all year while the agent's real score slides from 83% to 60%, how to answer the ROI question, and why containment targets differ by industry in Druid's 2026 AI Adoption Benchmark. In this talk: • Why containment rate can reward the wrong agent • A seven-pillar scorecard: configuration, runtime and governance • Scoring and weighting 26 components by business impact • Answering ROI and setting industry-specific targets SPEAKER Dan Bălăceanu, Co-founder & Chief Product Officer, Druid AI LinkedIn: https://www.linkedin.com/in/dan-balaceanu/ LINKS Druid AI: https://www.druidai.com Druid 2026 AI Adoption Benchmark: https://www.druidai.com/ai-adoption-benchmark CHAPTERS 0:00 Intro 1:36 Two interns, one dashboard 3:09 The containment trap 3:33 The seven-pillar framework 5:42 Scoring: 0, 1 or 2 7:12 Weighting by business impact 8:56 Governance: tolerance and risk 11:30 The agent's mandate 13:25 Platform 16:07 Validating outcomes 16:26 The final score 17:25 Recap 18:20 Green dashboard, falling score 18:50 The ROI question 20:00 Targets differ by industry Recorded at the AI Engineer World's Fair 2026 in San Francisco. Subscribe for more talks from the engineers building with AI. AI Engineer: https://ai.engineer YouTube: https://www.youtube.com/@aiDotEngineer X: https://x.com/aiDotEngineer LinkedIn: https://www.linkedin.com/company/aidotengineer/ #AIAgents #EnterpriseAI #AIEngineer canonical: >- https://liquidgold.cc/channel/aidotengineer/explore/video/containment-is-lying-to-you-measuring-ai-agents-lik-d4197f20ab7f markdown: >- https://liquidgold.cc/channel/aidotengineer/video/containment-is-lying-to-you-measuring-ai-agents-lik-d4197f20ab7f.md llms: null parent: 'https://liquidgold.cc/channel/aidotengineer.md' channel: aidotengineer updated: '2026-10-23T22:19:02.598Z' language: null children: [] videoId: PwCSYy-oGsM durationSeconds: 1320 publishedAt: '2026-10-12T00:00:10.000Z' searchChunks: - ordinal: 0 text: null startSeconds: 23 endSeconds: 103 watch: 'https://liquidgold.cc/watch/PwCSYy-oGsM?t=23' heldReason: rights - ordinal: 1 text: null startSeconds: 103 endSeconds: 115 watch: 'https://liquidgold.cc/watch/PwCSYy-oGsM?t=103' heldReason: rights - ordinal: 2 text: null startSeconds: 115 endSeconds: 197 watch: 'https://liquidgold.cc/watch/PwCSYy-oGsM?t=115' heldReason: rights - ordinal: 3 text: null startSeconds: 197 endSeconds: 207 watch: 'https://liquidgold.cc/watch/PwCSYy-oGsM?t=197' heldReason: rights - ordinal: 4 text: null startSeconds: 207 endSeconds: 218 watch: 'https://liquidgold.cc/watch/PwCSYy-oGsM?t=207' heldReason: rights - ordinal: 5 text: null startSeconds: 218 endSeconds: 228 watch: 'https://liquidgold.cc/watch/PwCSYy-oGsM?t=218' heldReason: rights - ordinal: 6 text: null startSeconds: 228 endSeconds: 345 watch: 'https://liquidgold.cc/watch/PwCSYy-oGsM?t=228' heldReason: rights - ordinal: 7 text: null startSeconds: 345 endSeconds: 355 watch: 'https://liquidgold.cc/watch/PwCSYy-oGsM?t=345' heldReason: rights - ordinal: 8 text: null startSeconds: 355 endSeconds: 439 watch: 'https://liquidgold.cc/watch/PwCSYy-oGsM?t=355' heldReason: rights - ordinal: 9 text: null startSeconds: 439 endSeconds: 451 watch: 'https://liquidgold.cc/watch/PwCSYy-oGsM?t=439' heldReason: rights - ordinal: 10 text: null startSeconds: 451 endSeconds: 538 watch: 'https://liquidgold.cc/watch/PwCSYy-oGsM?t=451' heldReason: rights - ordinal: 11 text: null startSeconds: 538 endSeconds: 550 watch: 'https://liquidgold.cc/watch/PwCSYy-oGsM?t=538' heldReason: rights - ordinal: 12 text: null startSeconds: 550 endSeconds: 698 watch: 'https://liquidgold.cc/watch/PwCSYy-oGsM?t=550' heldReason: rights - ordinal: 13 text: null startSeconds: 698 endSeconds: 711 watch: 'https://liquidgold.cc/watch/PwCSYy-oGsM?t=698' heldReason: rights - ordinal: 14 text: null startSeconds: 711 endSeconds: 809 watch: 'https://liquidgold.cc/watch/PwCSYy-oGsM?t=711' heldReason: rights --- # Containment Is Lying to You: Measuring AI Agents Like Employees — Druid AI Two AI interns. One handles 96% of conversations alone, the other only 80%. The dashboard picks the wrong one. Druid AI's team, including co-founder and chief product officer Dan Bălăceanu, shows why containment rate, the metric on every agent dashboard, can point you to the wrong agent. In their bank example, the intern who escalated more often was handing senior bankers customers who were about to refinance. They walk through a performance review framework built with their customers: seven pillars split into what the agent is configured with and what it does at runtime, with governance across both. Twenty-six components are each scored 0, 1 or 2 (missing, defined, or actually governed and tracked over time) and weighted by business impact, with governance and mandate weighted highest. They cover error tolerance, risk, compliance, observability, drift, hallucination, tone and sentiment, the agent's mandate, scope, business case and budget, a shared platform, and validating outcomes. They also show a dashboard that stays green all year while the agent's real score slides from 83% to 60%, how to answer the ROI question, and why containment targets differ by industry in Druid's 2026 AI Adoption Benchmark. In this talk: • Why containment rate can reward the wrong agent • A seven-pillar scorecard: configuration, runtime and governance • Scoring and weighting 26 components by business impact • Answering ROI and setting industry-specific targets SPEAKER Dan Bălăceanu, Co-founder & Chief Product Officer, Druid AI LinkedIn: https://www.linkedin.com/in/dan-balaceanu/ LINKS Druid AI: https://www.druidai.com Druid 2026 AI Adoption Benchmark: https://www.druidai.com/ai-adoption-benchmark CHAPTERS 0:00 Intro 1:36 Two interns, one dashboard 3:09 The containment trap 3:33 The seven-pillar framework 5:42 Scoring: 0, 1 or 2 7:12 Weighting by business impact 8:56 Governance: tolerance and risk 11:30 The agent's mandate 13:25 Platform 16:07 Validating outcomes 16:26 The final score 17:25 Recap 18:20 Green dashboard, falling score 18:50 The ROI question 20:00 Targets differ by industry Recorded at the AI Engineer World's Fair 2026 in San Francisco. Subscribe for more talks from the engineers building with AI. AI Engineer: https://ai.engineer YouTube: https://www.youtube.com/@aiDotEngineer X: https://x.com/aiDotEngineer LinkedIn: https://www.linkedin.com/company/aidotengineer/ #AIAgents #EnterpriseAI #AIEngineer Canonical: https://liquidgold.cc/channel/aidotengineer/explore/video/containment-is-lying-to-you-measuring-ai-agents-lik-d4197f20ab7f · Channel: [AI Engineer](https://liquidgold.cc/channel/aidotengineer.md) > Search chunk text is held until public-source provenance is verified. ## Document chunks --- #### Chunk 0: Document excerpt (0:12 to 1:43) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 1: Document excerpt (1:43 to 3:38) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 2: Document excerpt (3:38 to 5:45) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 3: Document excerpt (5:45 to 8:58) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 4: Document excerpt (8:58 to 11:38) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 5: Document excerpt (11:38 to 13:29) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 6: Document excerpt (13:29 to 16:07) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 7: Document excerpt (16:07 to 17:28) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 8: Document excerpt (17:28 to 18:32) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 9: Document excerpt (18:32 to 20:08) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 10: Document excerpt (20:08 to 22:00) > Held: transcript rights do not permit this chunk's text.