The Agentic Age

Ruth had printed the video comments. Not all of them. Just enough to fit on three pages.

Maren found the pages on the conference table between two financial models, clipped neatly at the corner. The first comment began halfway down the page.

Below it:

Ruth sat across from her with the Marcus Bell case open on one screen and the savings model open on the other. She touched neither.

"I watched it," she said.

Maren put her notebook on the table. "I assumed you had."

"Three times."

"Why three?"

"The first time I watched the video. The second time I watched the sequence. The third time I watched for what the first two had made me miss."

Maren sat down. "And?"

Ruth looked at the printed comments. "I don't like it."

"That isn't a finance answer."

"No."

"Do you've one?"

"That depends on what you're asking."

Maren opened her notebook. "I'm asking what the model sees when a case like Marcus Bell's happens."

"Nothing."

The answer came so quickly that Maren stopped writing.

Ruth turned the savings screen toward her. "The case closed on time. No rework. No formal complaint through the client channel. No service credit issued. No additional labor recorded. No escalation. The survey was sent and not completed."

"The post?"

"Not in the model."

"The reputational exposure?"

"Not in the model."

"The client risk?"

"Not yet in the model."

Maren looked at her. "Yet."

"If a client leaves, the loss is measurable. If a client asks for a concession, the concession is measurable. If the firm has to add labor back into the process, the labor is measurable."

"And before that?"

Ruth folded her hands. "Before that, you're asking me to price something that hasn't entered a financial system."

"That doesn't mean it isn't a cost."

"I didn't say it did."

The room was quiet. Ruth reached for the printed comments and squared the pages against the table.

"That's the part you keep waiting for me to argue with," she said.

"I'm not waiting for you to argue."

"You're a little."

Maren looked down at her notebook.

"The savings are real," Ruth continued.

"I know."

"The capacity gain is real."

"I know."

"The cycle-time reduction is real."

"I know that too."

"And this case is also real."

Maren looked up. Ruth's expression had not changed.

"I can show Gerald the first three to the dollar," Ruth said. "I can't show him the fourth one that way."

"Can you show him a range?"

"Based on what?"

"Reputational risk. Retention. Client exposure."

"Which client has left?"

"None that we know of."

"Which contract has changed?"

"None."

"Which employee claim is tied to this?"

"None."

"Then, no." Ruth said it without satisfaction. "I'm not going to invent a number because the absence of one is inconvenient."

"I didn't ask you to invent one."

"You were about to."

Maren closed her mouth.

Ruth looked again at the Marcus record. "I believe you that something is wrong," she said. "That isn't the same as being able to put it in this model." She tapped the savings screen once. "And Gerald lives in this model."

· · ·

The video was ten days old, and Maren's second month was ending. She had come to Ruth to test the finding before she took it to Gerald. That was what she told herself. Testing was responsible. Good practitioners tested. They tried to break their own conclusions before someone else did it for them in a room with more consequences.

She had expected Ruth to challenge the facts. Instead, Ruth had accepted them, which left Maren with a harder problem. The financial case was not false. It was incomplete in a way that finance had no instrument for correcting.

Maren went back to the borrowed office and put the two versions of the story side by side.

On the left:

On the right:

She stared at the page. Then she added Priya.

Then Desmond.

Then the manual-touch trend. The touches fell after each expansion. The implementation log called the fall progress.

None of it disproved the numbers on the left. That was the difficulty. The two columns did not contradict each other. They simply did not belong to the same instrument.

Before she drafted anything, she went back to the beginning. The original change management plan was on the shared drive where Daniel's team had left it, version 3.2, last modified 11 days before go-live.

It was good work. She read it the way she read every inherited plan, looking for the corner that had been cut, and did not find one. Audience analysis, impact assessment, communication plan, training plan, manager toolkit, adoption targets. The success criteria sat on the fourth page. Training completion above ninety percent. Usage at thirty days. Satisfaction at sixty. Escalation volume within threshold. Every criterion had been met, and the plan had been marked closed at the end of hypercare, sixty days after launch.

Nothing in it had been touched since. The agent had been expanded twice after the plan closed, into new case types, with new templates and a new routing rule. The document described an operation that no longer existed, and its last version note read Final.

She started a new document. At the top, she typed:

Then:

She read it. Accurate. Careful. Almost useless.

She deleted insufficient to establish sustained quality and tried again.

Better. She stopped. The sentence assumed that the human review activity should have remained. She believed that now. Believing it was not the same as proving the scale of the consequence.

She added:

Maren sat back. There it was. Not elegant. Not soft either.

Her phone buzzed. Gerald.

Maren looked at the time. 3:17. She had forty-three minutes to decide what she meant by read.

· · ·

At 3:24, Daniel appeared in her doorway. He knocked on the frame even though he was already standing inside it.

"Bad time?"

Maren minimized the document. "Depends."

"Gerald asked me for the expansion materials."

"Tomorrow's meeting?"

Daniel nodded. "He wants the next three functions mapped against the same rollout sequence. I told him I'd wait until he talked to you."

"That was considerate."

"I wasn't being considerate." He smiled.

Daniel always looked slightly relieved to be talking to her, as if every conversation transferred a small amount of responsibility across the table.

"I need to know whether you're changing the playbook," he said.

"I don't know yet."

"That isn't the answer I was hoping for."

"It’s the only one I've at the moment. Sorry."

He leaned against the doorframe. "Is this about the post?"

"Yes."

"One case?"

Maren looked at him. Daniel lifted a hand. "I know. That sounded worse out loud."

"It did."

"I'm not dismissing it."

"No."

"I'm trying to understand whether we're dealing with an incident or a program finding."

"So am I. When was the change plan last revised?" Maren asked.

Daniel blinked at the turn. "Revised?"

"Updated. Reopened. Anything."

"It closed out with hypercare. Everything transitioned to the working team." He said it without defensiveness, the way a person reports that winter was followed by spring.

"Who owns the plan now?"

"The working team."

"The working team isn't a person."

"No," Daniel said, agreeably. "It's a team."

He glanced toward her screen. "You have been asking for records that weren't in the original assessment list."

Maren said nothing.

"And Tomas said you pulled old manual-touch data."

"Did he?"

"He wasn't complaining."

"OK."

"I'm not either."

Maren waited. Daniel looked toward the hall, then back at her.

"People have noticed that the questions changed."

There it was. Not a warning, exactly, but the shape of one arriving early.

"Which people?"

"People who have calendars."

Maren gave a small laugh. "Very specific."

"You know how this works."

"I do."

"I thought you might." Daniel pushed away from the frame. "For what it's worth, Gerald meant what he said on the first day. He really did bring you here because he wanted someone outside to tell him whether this holds."

Maren looked at him.

"I'm not sure anyone thought about what happens if you say it doesn't," Daniel added. Then he left.

· · ·

At 3:52, Maren printed the Marcus sequence. At 3:54, she printed the manual-touch trend. At 3:56, she removed both from the folder. At 3:57, she put them back. At 3:59, she took the elevator one floor up.

Gerald's office was larger than she expected and less impressive than it had any reason to be. A round table. Two chairs near the window. A bookshelf with real books among the awards. A framed photograph of Gerald standing beside a group of employees in matching volunteer shirts. No desk between them. Of course.

Gerald stood when she came in. "Maren. Right on time." He closed the door himself. "Coffee?"

"No, thank you."

"Good judgment. I've had enough for both of us." He gestured toward the round table.

A board packet was already there. Maren recognized the blue ribbon from the lobby screen on the cover.

Gerald saw her looking at it. "Tomorrow isn't an approval meeting," he said.

"No?"

"It's a preparation meeting for the approval meeting."

"Important distinction."

"I thought you'd appreciate it."

She sat. Gerald took the chair across from her. "I want your honest read."

Maren put the folder on the table. "You may need to be more specific."

He smiled. "Fair." He opened the board packet. "We're not asking you to bless the technology. We're asking whether the way we drove adoption here can travel. Three more functions. Similar administrative load. Similar opportunity."

"Similar work?"

"Not identical."

"Similar risk?"

"Probably not."

He nodded as if she had confirmed something useful. "Exactly. That's why I want your read before we get too far."

Maren looked at the packet. The scale recommendation had her firm's name on the draft cover. Not her recommendation. Her name was there because she was the outside validation.

"You were going to put this in front of the board before my final report?"

"The packet is in development."

"My name is on it."

"Your engagement is on it."

"That isn't much of a distinction from where I'm sitting."

Gerald closed the packet. "Then let's not get distracted by the packet."

Maren almost laughed. Gerald did. "Fine," he said. "Bad sentence." He folded his hands. "What did you find?"

Maren opened the folder. She started with Marcus. Not the theory. Not the dashboard. The man.

She put the printed sequence on the table and turned it toward Gerald. He read the first line. His face changed. Not dramatically. Enough. He read the whole page before he spoke.

"I saw the video."

"I assumed."

"This is worse on paper."

"Yes."

Gerald leaned back. "Has anyone spoken to him?"

"Client services has reached out."

"I mean us."

"I don't know."

Gerald picked up the sequence again. "Congratulations." He said the word quietly. Maren waited. He put the page down. "This shouldn't have happened."

"No."

"What failed?"

"The case didn't fail."

Gerald looked at her.

"The sequence ran correctly. The classification held. The workflow closed on time. No rule triggered an escalation."

"I understand that. I'm asking what failed."

Maren took out the manual-touch trend. "This used to be caught by people."

"Marcus specifically?"

"No."

"Then help me."

"Before the expansion, specialists were editing more first messages. They were changing tone, swapping sequences, adding context. Those touches fell after each expansion."

"Because people trusted the tool."

"That was how the decline was interpreted."

"Do you disagree?"

"I think trust is one possible explanation."

"What's the other?"

"That we removed the activity that was catching what the rules couldn't see."

Gerald reached for the sequence again. "Who approved this template?"

"It predates the agent. It's the firm's language. It's in the brochures."

He was quiet for a moment, and Maren understood that he had recognized the sentence.

"Then why did it go out first? To a man we knew was being reduced."

"The routing was correct. The content was approved. What the workflow lost was the person who decided when a message like that could go to whom. The specialists weren't fixing broken templates. They were choosing moments. That was never written down anywhere, so nothing noticed when it stopped."

Gerald put the page down and looked at the graph.

"How many cases?"

"We don't know."

"How many failures?"

"We don't know that either."

"How often did these manual touches improve a case versus simply reflecting people checking work the system had already done correctly?"

"That data isn't available either."

Gerald put the graph down. Maren felt her language narrowing.

"Those aren't the same unknown."

"I didn't say they were."

"The first is a measurement gap. The second is a quality gap."

"Maren."

"The system can't distinguish between a case that never needed human judgment and a case that succeeded because someone supplied judgment before the system could record the absence of it."

Gerald watched her. There was no irritation in his face. That made it worse.

"I hear you," he said.

She almost stopped him, but chose to hold off on her follow-up statement.

"I hear the concern. I also need you to help me square it with what we know." He opened the board packet and slid out a single page. The same figures from the lobby. Cycle time. Capacity. Satisfaction. Escalations.

"The numbers are very good," he said. "They're the best numbers this function has ever posted."

"I know."

"Ruth tells me they hold."

"They do."

"The specialists are using the agent."

"Yes."

"They like it."

"Yes."

"Client renewals have held."

"So far."

"No service-level penalties."

"No."

"No unusual rework."

"Not that we can see."

Gerald paused. "Do you hear yourself?"

Maren did. She hated that she did.

He did not press the point. "I have one terrible case," Gerald said. "I have a public embarrassment that I'm not minimizing. I also have an operation handling thousands of transitions faster, with less overtime, and with satisfaction holding."

"The measures would have scored Marcus as satisfied if he had completed the survey."

"But he didn't."

"That isn't the point."

"Then tell me the point in a way I can act on."

The room went quiet. Maren looked at the sequence on the table. The line was still there.

"The point is that the success framework can't see the failure," she said.

Gerald waited. "That's a strong sentence."

"It's an accurate one."

"Is it?"

"Yes."

"Or is it the hypothesis you're testing?"

Maren looked at him. He was not trying to trap her. He genuinely wanted the distinction. That was the problem.

"The Marcus case proves the framework missed Marcus."

"Yes."

"The manual-touch data shows that the review behavior that used to catch context-sensitive cases declined."

"Yes."

"Priya and Desmond both raised concerns before closeout."

Gerald's eyes shifted slightly. "Desmond."

"Yes."

"I've read Desmond's notes."

"Have you?"

"I've read summaries of them."

Maren said nothing. Gerald caught it.

"That's fair," he said.

"No, it isn't."

He looked at her. She had not meant to say it that quickly. Maren reached for her notebook, then stopped herself. Gerald noticed that too.

"Go on."

"You didn't read his concern. You read what survived the chain."

The room changed, the way a room does when a true thing has been said plainly and cannot be taken back. Gerald sat back. Maren could feel the next sentence waiting. She could also feel how easily it could become the wrong sentence.

"I'm not saying anyone hid it from you," she said.

"Good."

"I'm saying the concern changed shape every time it moved upward."

"That's possible."

"It became a training note. Then a manager reminder. Then a future governance item."

"Which may have been the correct response to what people knew at the time."

"Yes."

Gerald leaned forward. "Then what exactly are you asking me to accept?"

Maren looked at the board packet. There were several answers. Pause expansion. Reopen sampling. Restore qualitative review. Reconstruct the cases that would once have received human touches. Reopen a change plan that had been closed on launch numbers and never revised. Build measures for judgment, not only usage.

None of those was the real question. The real question was whether Gerald was willing to accept that the victory he had already carried to the board might contain a failure his own evidence could not disprove.

Maren did not say that. Not yet.

"I'm asking you not to scale this model tomorrow."

Gerald's face went still. "The meeting tomorrow isn't an approval."

"You know what I mean."

"Yes." He looked at the graph again. "For how long?"

"I don't know yet."

"Until what?"

"Until we can tell the difference between routine cases and cases the system only thinks are routine."

"How?"

"I'm working on that."

Gerald nodded once. "On what basis do I pause?"

"The basis is that we don't know what the current measures are missing."

He was quiet. Then: "Not knowing isn't evidence that the rollout failed."

Maren looked at him. "No."

"And one case isn't evidence that the model fails at scale."

"No."

"And the decline in manual touches was an intended outcome."

"Yes."

"And the people doing the work supported that outcome."

"Yes."

Gerald spread one hand over the pages between them. "Then you're asking me to interrupt a successful operation because the very absence of evidence might be evidence."

Maren did not answer. He had made it sound absurd, but he had not made it inaccurate.

· · ·

Gerald stood and went to the window. Maren remained at the table. Outside, the parking structure blocked most of the view. Beyond it, a strip of sky had turned the dull white that came before rain.

Gerald kept his back to her. "I want to be careful here."

"So do I."

"I know." He turned. "That's why I hired you."

The sentence landed harder than it should have.

Gerald came back to the table. "I don't need you to tell me everything is fine."

"I know."

"I meant what I said on your first day. If something isn't solid, I want to hear it from you before I hear it from a client."

Maren looked at the Marcus sequence. "Unfortunately, we heard it from Marcus."

Gerald stopped. Maren continued before she could soften the sentence.

"He went public because it was the only avenue he had left. Not to report the failure. To get someone to answer for it. Inside the firm, there was nothing to answer; the case had closed, the survey had gone out, and nothing in the process had called it wrong."

She let that sit. "The only place that registered it as wrong was the internet."

Gerald's eyes stayed on her. "That isn't the same as saying the entire rollout is a failure."

"I didn't say it was."

"You're asking me to stop scaling it."

"I'm asking you not to replicate what we don't yet understand."

The room was quiet again. Gerald sat. When he spoke, his voice was softer. "What would you put in the report today?"

Maren looked at him. "Today?"

"Right now. No more investigation. No better data. No additional interviews. You have to write the findings with what you know now."

Maren did not answer immediately. That was the question she had been avoiding since she pulled the Marcus Bell record, though she would not have described it that way.

She opened her folder and pulled out the page she had written an hour earlier. Gerald read it.

He looked up. "This I agree with."

Maren said nothing. He read the second sentence.

He read it twice. "This one I'd ask you to prove."

"I expected that."

"Can you?"

"Not yet."

He moved to the third.

Gerald stopped there. The silence lengthened. Finally he said, "That one stays."

Maren had not expected that.

Gerald put the page down. "I'm not fighting you on Marcus."

"This isn't about fighting."

"No. It's about what follows." He turned the page around so it faced her. "Here is what I can support today. We acknowledge the case. We add qualitative sampling. We restore a human review requirement for selected transition profiles. We strengthen escalation guidance. We don't present the current dashboard as sufficient evidence of quality."

Maren listened.

"And we keep tomorrow's meeting," Gerald continued.

"There it is."

"Yes."

"You want to treat this as optimization."

"I want to treat what you can prove as what you can prove."

"The model may be producing failures we can't see."

"Maybe."

"The review behavior that used to catch them is gone."

"Mostly gone."

"Because we celebrated its disappearance."

Gerald's jaw tightened for the first time. "Maren."

She stopped. Not because he had raised his voice. He had not. Because he had finally used the tone of someone reminding another person where the room ended.

Gerald let the moment settle. Then his voice returned to normal. "You're very close to making a larger claim than your evidence can carry."

Maren looked at the page. He was right.

That was the wall. Not denial. Not cowardice. Evidence. True numbers. A real result. And the professional discipline not to say more than you could prove.

Gerald slid the page back to her. "Write the finding you have." Maren did not move. "Do not write the one you're afraid might be true."

The sentence stayed between them.

Gerald glanced at the board packet. "I'll take your name off tomorrow's materials."

Maren looked up. "Why?"

"Because your validation isn't complete."

"That isn't what you said fifteen minutes ago."

"Fifteen minutes ago, I thought I understood what you were validating."

There was no anger in it. Only adjustment. The first clean consequence.

Gerald closed the packet. "You have through Friday for the interim report. I need a recommendation I can act on."

"And if my recommendation isn't to scale?"

"Then write that." He met her eyes. "But be prepared to defend it against the facts we have, not the facts you think might appear later."

· · ·

Maren left Gerald's office carrying the same folder she had brought in. It felt heavier.

The elevator opened on her floor. Daniel was near the reception desk, phone in hand. He looked up when she stepped out. Neither of them spoke. Whatever he saw in her face, it was not the answer he had been hoping for, and he did not look surprised that it wasn't.

Maren walked past him. At her borrowed office, she closed the door and put the folder on the table.

Her original scope statement was still in the project folder. She opened it.

She had read the line before. Many times. Now it looked different.

Not because Gerald had threatened her. He had not. Not because he had dismissed the finding. He had kept the hardest sentence. Not because he had asked her to lie. He had asked her to prove what she wanted to say.

Maren opened the interim report. The first sentence waited.

She left it. The second:

She highlighted it. Her finger hovered over delete.

She thought of Gerald saying, Write the finding you have. She thought of Desmond's third memo. Post-Go-Live Stabilization Considerations. The concern softened until it could survive the room.

Maren took her hand away from the keyboard. She opened a second document. At the top, she typed:

Then:

She read it once. Then again. The sentence would cost more.

She saved both documents.

For the first time since the engagement began, Maren knew exactly which one Gerald would rather receive. She also knew which one was true.