# The Agent Works. Does Anyone Trust It? Michele Grant · AI Engineer NYC 2026 · October 14 [Watch recording](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16107s) [04:28:27](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16107s) GoLive proves that a system exists, but it doesn't prove the people behind the work will actually use it. [04:28:35](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16115s) or that it will ever deliver a single dollar from the deal model. [04:28:39](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16119s) I am Michele Grant, creator of Leading at Scale, [04:28:43](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16123s) and the question every enterprise is confronting right now is the agent works. [04:28:49](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16129s) Does anyone trust it? [04:28:51](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16131s) I have spent over 20 years in enterprise transformation [04:28:55](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16135s) and more than a decade leading M&A integrations across media, tech, and consumer goods. [04:29:03](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16143s) On those deals, trust was not something we crossed our fingers and hoped for after launch. [04:29:11](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16151s) We engineered it before day one through early change management and risk management. [04:29:19](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16159s) So when the platform went live, the teams on the ground actually relied on it. [04:29:27](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16167s) AI agents demand the exact same muscle. [04:29:31](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16171s) An agent earns trust when it delivers the value the business budgeted for while staying [04:29:39](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16179s) safely within the company's approved risk appetite, policies, and regulatory guardrails. [04:29:47](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16187s) In my experience, leadership gets there by answering three straightforward questions. [04:29:54](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16194s) What value does it create? [04:29:57](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16197s) Who is on the hook for it? [04:29:59](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16199s) And [04:30:00](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16200s) And what evidence gives it permission to run with greater autonomy? [04:30:05](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16205s) The moment an agent steps out of a sandbox and into a core business workflow, the entire test changes. [04:30:14](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16214s) A pilot tests whether the software functions, but daily operations is where we test whether it creates value within the controls set by leadership. [04:30:27](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16227s) Established AI risk frameworks give you strong principles. [04:30:31](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16231s) What they do not give you is how to apply these principles to each agent. [04:30:37](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16237s) For every agent, leadership has to decide four things. [04:30:41](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16241s) What is it permitted to do? [04:30:44](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16244s) Who owns it? [04:30:45](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16245s) What evidence it records? [04:30:48](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16248s) And what triggers an escalation? [04:30:52](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16252s) That work is not an engineering task. [04:30:55](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16255s) We find it is an executive operating mandate. [04:30:59](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16259s) And today, I'll show you how to meet it. [04:31:03](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16263s) Let us ground this in a real-world scenario. [04:31:07](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16267s) own. Picture a $15 billion acquisition targeting 400 million in net synergies. The deal has already [04:31:17](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16277s) closed and the integration efforts are now underway. The integration management office, [04:31:23](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16283s) which we'll also refer to as the IMO today, is driving the synergy roadmap across 40 work streams, [04:31:32](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16292s) covering everything from rationalizing vendor contracts to consolidating ERP systems. [04:31:39](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16299s) And every single initiative has a business owner accountable for its bottom line savings. [04:31:47](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16307s) Every week, analysts in the IMO are going to reconcile the numbers against the general ledger [04:31:56](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16316s) as they track down status from all 40 leads. [04:32:01](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16321s) They're also comparing actuals against the plan, [04:32:05](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16325s) then finally assembling the executive deck for the steering committee, [04:32:10](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16330s) which is going to include the CFO, the chief risk officer, [04:32:15](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16335s) business unit heads, and the integration lead. [04:32:19](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16339s) The engineering team builds an agent to automate that first draft. [04:32:26](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16346s) The agent then ingests the weekly updates, reconciles the ledger, runs variance math against the plan, and flags the initiatives that are falling behind. [04:32:38](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16358s) Everything looks good, and they decide to run a four-week pilot. [04:32:42](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16362s) The agent's drafts match the analyst's reports, line for line. [04:32:48](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16368s) So we can say that the tech works. [04:32:51](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16371s) Now, the question is, what value does that steering committee need before they can trust it to run the business? [04:33:01](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16381s) First, the steering committee needs a written value case, agreed upon before the agent ever touches production data. [04:33:11](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16391s) In our example, that comes down to three specific metrics. [04:33:17](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16397s) First, let's look at cycle time. [04:33:20](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16400s) If that weekly report eats five business days to compile right now, you're going to want to set a hard target such as two days. [04:33:30](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16410s) Your second metric will be analyst capacity. [04:33:34](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16414s) If four analysts spend half of their week building the stack, that is 80 hours a week the agent gives back. [04:33:44](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16424s) That time counts as value only when the integration lead assigns it to the work streams that are behind on their synergy targets and need help getting back on track. [04:33:57](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16437s) Then you only count that value after netting out the compute, licensing, and human review costs of running the agent. [04:34:09](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16449s) Your third metric is the most important one, detection time. [04:34:14](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16454s) That is how long it takes from the first sign of a slip in the data to the moment the steering committee sees it. [04:34:21](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16461s) Cutting the report from five days to two days does not guarantee that the committee sees that slip any sooner. [04:34:30](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16470s) The agent has to catch it, and the value case measures that directly. [04:34:37](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16477s) Every single metric requires a pre-deployment baseline. [04:34:41](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16481s) If the IMO cannot state the baseline today, do not deploy. [04:34:47](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16487s) If you can't measure where you started, you can never prove the agent moved the needle. [04:34:54](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16494s) Next, the steering committee needs to know who owns the risk the agent creates. [04:35:01](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16501s) We are going to apply the three-line model to the agent. [04:35:06](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16506s) In the first line, management owns the controls. [04:35:10](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16510s) The integration lead is the accountable executive. [04:35:14](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16514s) She owns the agent's output exactly the way she owned the manual deck. [04:35:20](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16520s) In the second line, risk and compliance oversee the agent and challenge how to run it. [04:35:28](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16528s) How deep they go depends on the agent's exposure. [04:35:32](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16532s) An internal draft scrubbed by a senior analyst warrants a very different review standard than an autonomous agent publishing figures straight to the board. [04:35:44](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16544s) In the third line, internal audit provides assurance deciding based on material risk when and how the agent enters the annual audit plan. [04:35:56](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16556s) The agent also needs one employee assigned to oversee it every day. [04:36:04](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16564s) In our example, that is the senior analyst. [04:36:08](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16568s) reviews the draft before it reaches the community. [04:36:12](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16572s) Her role requires four specific capabilities. [04:36:15](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16575s) She has to first understand the operational boundaries. [04:36:20](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16580s) She has to interpret its output with context. [04:36:23](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16583s) She must have the authority to override or reverse it, and she needs a clear kill switch to shut it down safely the second it behaves unexpectedly. [04:36:39](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16599s) Accountability also means defining where the agent's authority stops. [04:36:44](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16604s) I document this in a one-page policy called the Human Works Standard. [04:36:50](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16610s) For every fork in the process, it maps the decision the human role authorized to make it, the required evidence, and the escalation path. [04:37:03](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16623s) These decisions exist at every tier. [04:37:07](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16627s) senior analyst decides whether a variance is an accounting timing difference or a real cash shortfall. [04:37:14](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16634s) The work stream lead decides how to remediate that shortfall. [04:37:19](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16639s) And the integration lead is going to decide whether to formally flag an initiative as at risk to the steering committee. [04:37:28](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16648s) The agent prepares the analysis behind every one of those decisions, and it stops there. [04:37:37](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16657s) The person who makes the decision owns the consequence. [04:37:42](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16662s) So keeping these decisions with your team does not slow the agent down. [04:37:47](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16667s) It can go into use while each role keeps its authority. [04:37:53](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16673s) one reason to delay deployments is a decision with no assigned owner. [04:38:00](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16680s) This means if leadership has not agreed who declares an initiative off track, [04:38:07](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16687s) the agent's flags lead to no action. The third question is where the engineers in this room [04:38:17](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16697s) come in. The evidence for greater autonomy comes from how the agent is built. I look for three [04:38:26](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16706s) artifacts. The first is an auditable decision log. For every work stream the agent evaluates, [04:38:34](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16714s) it records the raw data ingested, the tool calls executed, the prompt and the model versions, [04:38:43](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16723s) output generated along with the human sign-off. Two engineering choices make that log airtight. [04:38:53](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16733s) First, your core math and reconciliation rules must run in deterministic finance-approved code, [04:39:02](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16742s) not generated as conversational prose from an LLM. [04:39:08](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16748s) Second, treat the model's written summary as a helpful explanation, not proof. [04:39:14](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16754s) Your real proof lives in the telemetry. [04:39:18](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16758s) What records were pulled? [04:39:20](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16760s) What deterministic tools ran? [04:39:23](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16763s) And who signed the review? [04:39:27](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16767s) If reported savings do not match the general ledger, [04:39:31](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16771s) Agent does not touch the number. [04:39:35](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16775s) flats discrepancy and roots it up. [04:39:39](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16779s) second artifact is the model inventory record. [04:39:44](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16784s) captures the accountable executive. [04:39:48](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16788s) lineage. [04:39:50](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16790s) Dependencies across source systems. [04:39:53](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16793s) Validation results. [04:39:55](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16795s) Escalation path. [04:39:57](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16797s) And the kill switch runbook. [04:39:59](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16799s) Most [04:40:00](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16800s) Most importantly, it documents your regression triggers, which are any updates to our prompt, base model, or source API that require full revalidation before running in production again. [04:40:16](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16816s) third artifact, this is going to be the Autonomy Scorecard. [04:40:21](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16821s) The steering committee uses it to grant autonomy in four stages. [04:40:27](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16827s) Stage one is shadow mode. [04:40:29](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16829s) The analysts build the report as they do today. [04:40:34](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16834s) agent runs right alongside them and the team compares the two. [04:40:38](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16838s) Stage two is the supervised draft. [04:40:42](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16842s) The agent compiles the first draft and the senior analyst reviews every single line item. [04:40:50](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16850s) Stage 3 is the exception and sample, where the analyst reviews land initiatives plus a random sample of the claimants. [04:41:02](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16862s) Then we have stage 4, where it's exceptions only. [04:41:07](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16867s) The analyst reviews only high-risk anomalies, supported by periodic independent sampling. [04:41:14](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16874s) Pay close attention to stage 4, however, because once humans only look at exceptions, your biggest vulnerability is what the agent failed to flag. [04:41:27](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16887s) So I advise tracking missed material issues through independent sampling, monitoring error severity, and watching your override rate carefully. [04:41:38](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16898s) A falling override rate does not automatically mean the agent is getting smarter. [04:41:44](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16904s) It can also mean the depth of review has split. [04:41:48](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16908s) So independent sampling tells you which one of these you are seeing. [04:41:53](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16913s) The scorecard puts those control metrics side by side with your value metrics, [04:42:00](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16920s) cycle time, analyst hours deployed, and your detection time. [04:42:05](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16925s) And with that single page view, the steering committee can make clean, efficient data-backed decisions [04:42:12](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16932s) to either promote the agent, hold it where it is, step it back, or down to supervised review stage. [04:42:23](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16943s) So let's come back to the three questions. [04:42:28](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16948s) Ask them of every agent you build, fund or govern. [04:42:33](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16953s) What value does it create? [04:42:36](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16956s) Who is accountable for it? [04:42:38](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16958s) and what evidence permits it to operate with autonomy. [04:42:44](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16964s) In our M&A example today, the agent does not magically create a 400 million in synergies. [04:42:51](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16971s) Procurement, supply chain, and operating leaders capture that value through the gritty work of integration. [04:42:59](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16979s) what the agent delivers is clear. [04:43:02](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16982s) Verified evidence weeks ahead of schedule, [04:43:05](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16985s) giving leadership the runway to intervene while the deal's synergies can still be captured. [04:43:13](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16993s) Building an agent that works is just the first step. [04:43:17](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=16997s) Answering those three questions is what earns it the trust to run at scale. [04:43:24](https://www.youtube.com/watch?v=jiUC6IUMdvc&t=17004s) Thank you.