--- schema: 1 type: video title: >- Why Your Enterprise Tech Stack Isn’t Ready for AI Agents — Christopher Lovejoy & Saul Howard description: >- The proof of concept works. It hits the accuracy targets, it is fast, it is cheap, and the room is happy. Then someone from compliance raises a hand and asks to see the audit trail, and the whole thing stops. Christopher Lovejoy and Saul Howard have watched that meeting happen repeatedly, and their point is that an audit trail is not a developer log. Under the frameworks enterprises actually answer to, it is a complete record of every action an agent took, every place it touched data, and the authorization behind each one, durable enough to stand up as a chain of evidence if the decision were ever examined in court. Their answer is to take the constraints seriously first and rebuild toward the accuracy afterwards, rather than bolting requirements onto a demo. An immutable append only event log makes auditability fall out of the storage model instead of being reconstructed later, at the cost of harder reads. Patient data lives in schema driven object storage alongside that log rather than inside it, so the events hold only references, which lets engineers debug what an agent did without being exposed to the health data itself, and gives a natural place to enforce zero trust and constrain prompt injection. Escalation works because humans and models are both treated as agents, so any action either can take, the other can take too. Evaluation then emerges from those three primitives rather than being attached to the side, including on production data that never leaves the customer's environment. Speaker info: - https://x.com/ChrisLovejoy_ - https://www.chrislovejoy.me - https://x.com/saulhoward - https://linkedin.com/in/saulhoward Timestamps: 0:00 - Why healthcare is hard, and what transfers to other regulated work 1:57 - The enterprise proof of concept 2:49 - What the buildout actually connects to 3:39 - Everyone assumes the hard part is done 4:28 - The questions that arrive the next day 5:19 - An audit trail is not a developer log 7:03 - The immutable event log, and its tradeoff 8:46 - What shape healthcare data actually has 10:28 - Object storage beside the log, not inside it 11:20 - Debugging an agent without seeing the data 12:15 - Zero trust and the lethal trifecta 13:07 - Escalation when you cannot predict it 13:58 - Treating humans and models as the same kind of agent 14:49 - Why evals are hard here 15:39 - Evaluation as a byproduct of the primitives 17:20 - Architecture as choosing what stays simple 18:08 - Where it goes wrong, and where it goes right --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/mav15aW9lLM-why-your-enterprise-tech-stack-isn-t-ready-for-ai-agents-and-what-to-build-instead Join our upcoming AI Engineer conferences (see https://ai.engineer landing page): - AIEi Shanghai, November 5–6 — the first for meeting top Chinese labs and engineers - AIE CODE in SF, November 10–12 — the top AI Coding conference returns - AIEi Sydney, December 7–8 — the first AIEi run alongside NeurIPS 2026 in Sydney --- canonical: >- https://liquidgold.cc/channel/aidotengineer/explore/video/why-your-enterprise-tech-stack-isn-t-ready-for-ai-a-b395a8a2dda6 markdown: >- https://liquidgold.cc/channel/aidotengineer/video/why-your-enterprise-tech-stack-isn-t-ready-for-ai-a-b395a8a2dda6.md llms: null parent: 'https://liquidgold.cc/channel/aidotengineer.md' channel: aidotengineer updated: '2026-10-24T05:02:04.218Z' language: null children: [] videoId: mav15aW9lLM durationSeconds: 1155 publishedAt: '2026-08-19T18:30:15.000Z' searchChunks: - ordinal: 0 text: null startSeconds: 25 endSeconds: 126 watch: 'https://liquidgold.cc/watch/mav15aW9lLM?t=25' heldReason: rights - ordinal: 1 text: null startSeconds: 182 endSeconds: 227 watch: 'https://liquidgold.cc/watch/mav15aW9lLM?t=182' heldReason: rights - ordinal: 2 text: null startSeconds: 237 endSeconds: 270 watch: 'https://liquidgold.cc/watch/mav15aW9lLM?t=237' heldReason: rights - ordinal: 3 text: null startSeconds: 423 endSeconds: 434 watch: 'https://liquidgold.cc/watch/mav15aW9lLM?t=423' heldReason: rights - ordinal: 4 text: null startSeconds: 434 endSeconds: 526 watch: 'https://liquidgold.cc/watch/mav15aW9lLM?t=434' heldReason: rights - ordinal: 5 text: null startSeconds: 537 endSeconds: 639 watch: 'https://liquidgold.cc/watch/mav15aW9lLM?t=537' heldReason: rights - ordinal: 6 text: null startSeconds: 639 endSeconds: 650 watch: 'https://liquidgold.cc/watch/mav15aW9lLM?t=639' heldReason: rights - ordinal: 7 text: null startSeconds: 683 endSeconds: 695 watch: 'https://liquidgold.cc/watch/mav15aW9lLM?t=683' heldReason: rights - ordinal: 8 text: null startSeconds: 695 endSeconds: 739 watch: 'https://liquidgold.cc/watch/mav15aW9lLM?t=695' heldReason: rights - ordinal: 9 text: null startSeconds: 739 endSeconds: 751 watch: 'https://liquidgold.cc/watch/mav15aW9lLM?t=739' heldReason: rights - ordinal: 10 text: null startSeconds: 805 endSeconds: 849 watch: 'https://liquidgold.cc/watch/mav15aW9lLM?t=805' heldReason: rights - ordinal: 11 text: null startSeconds: 849 endSeconds: 859 watch: 'https://liquidgold.cc/watch/mav15aW9lLM?t=849' heldReason: rights - ordinal: 12 text: null startSeconds: 903 endSeconds: 939 watch: 'https://liquidgold.cc/watch/mav15aW9lLM?t=903' heldReason: rights - ordinal: 13 text: null startSeconds: 950 endSeconds: 1052 watch: 'https://liquidgold.cc/watch/mav15aW9lLM?t=950' heldReason: rights - ordinal: 14 text: null startSeconds: 1095 endSeconds: 1105 watch: 'https://liquidgold.cc/watch/mav15aW9lLM?t=1095' heldReason: rights - ordinal: 15 text: null startSeconds: 1105 endSeconds: 1155 watch: 'https://liquidgold.cc/watch/mav15aW9lLM?t=1105' heldReason: rights --- # Why Your Enterprise Tech Stack Isn’t Ready for AI Agents — Christopher Lovejoy & Saul Howard The proof of concept works. It hits the accuracy targets, it is fast, it is cheap, and the room is happy. Then someone from compliance raises a hand and asks to see the audit trail, and the whole thing stops. Christopher Lovejoy and Saul Howard have watched that meeting happen repeatedly, and their point is that an audit trail is not a developer log. Under the frameworks enterprises actually answer to, it is a complete record of every action an agent took, every place it touched data, and the authorization behind each one, durable enough to stand up as a chain of evidence if the decision were ever examined in court. Their answer is to take the constraints seriously first and rebuild toward the accuracy afterwards, rather than bolting requirements onto a demo. An immutable append only event log makes auditability fall out of the storage model instead of being reconstructed later, at the cost of harder reads. Patient data lives in schema driven object storage alongside that log rather than inside it, so the events hold only references, which lets engineers debug what an agent did without being exposed to the health data itself, and gives a natural place to enforce zero trust and constrain prompt injection. Escalation works because humans and models are both treated as agents, so any action either can take, the other can take too. Evaluation then emerges from those three primitives rather than being attached to the side, including on production data that never leaves the customer's environment. Speaker info: - https://x.com/ChrisLovejoy\_ - https://www.chrislovejoy.me - https://x.com/saulhoward - https://linkedin.com/in/saulhoward Timestamps: 0:00 - Why healthcare is hard, and what transfers to other regulated work 1:57 - The enterprise proof of concept 2:49 - What the buildout actually connects to 3:39 - Everyone assumes the hard part is done 4:28 - The questions that arrive the next day 5:19 - An audit trail is not a developer log 7:03 - The immutable event log, and its tradeoff 8:46 - What shape healthcare data actually has 10:28 - Object storage beside the log, not inside it 11:20 - Debugging an agent without seeing the data 12:15 - Zero trust and the lethal trifecta 13:07 - Escalation when you cannot predict it 13:58 - Treating humans and models as the same kind of agent 14:49 - Why evals are hard here 15:39 - Evaluation as a byproduct of the primitives 17:20 - Architecture as choosing what stays simple 18:08 - Where it goes wrong, and where it goes right --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/mav15aW9lLM-why-your-enterprise-tech-stack-isn-t-ready-for-ai-agents-and-what-to-build-instead Join our upcoming AI Engineer conferences (see https://ai.engineer landing page): - AIEi Shanghai, November 5–6 — the first for meeting top Chinese labs and engineers - AIE CODE in SF, November 10–12 — the top AI Coding conference returns - AIEi Sydney, December 7–8 — the first AIEi run alongside NeurIPS 2026 in Sydney --- Canonical: https://liquidgold.cc/channel/aidotengineer/explore/video/why-your-enterprise-tech-stack-isn-t-ready-for-ai-a-b395a8a2dda6 · Channel: [AI Engineer](https://liquidgold.cc/channel/aidotengineer.md) > Search chunk text is held until public-source provenance is verified. ## Document chunks > Missing: no stored document chunks are available.