The Recommendation Was Ready
Fiction
At 4:42, Mara was about to recommend a settlement she could defend line by line and could no longer explain as a whole.
The offer expired at five.
For eleven months, she had represented Bellwether Instruments in a contract dispute with its largest distributor. Depositions started Monday. Another month of litigation would consume money the company needed for manufacturing. The distributor's offer ended the damages claim, paid outstanding invoices at a discount, and terminated the distribution agreement.
On her screen, the recommendation was clear: accept.
The firm's AI tool had helped assemble the chronology, compare the drafts, and organize the litigation-cost scenarios. Mara's associate, Dev, had checked the underlying documents. Mara had reviewed the analysis herself. The dates matched. The release was broader than the first offer, and the draft flagged that. Nobody had invented a case.
That morning, she had told Dev exactly what would make the settlement acceptable: the payment mattered, but so did Bellwether's ability to keep servicing the instruments already installed at customers' sites.
Now the partner wanted a recommendation in three minutes.
“We can't send them another essay,” he said. “They're paying us to advise.”
He was right. The client had already complained about costs. Bellwether's general counsel, Anika, had a board call at five-thirty and needed something she could use.
Mara moved the cursor to the first sentence of her email.
We recommend accepting the revised terms.
Dev pointed at a paragraph halfway down the analysis. “Have they decided they can lose the service channel?”
The tool had identified the termination provision correctly. It had treated the loss of distributor support as a commercial consequence to be weighed against continued litigation expense. The summary described that consequence as manageable.
Mara had supplied that word.
Three days earlier, Anika had sent a restricted strategy note. A replacement distributor was interested, but nothing was signed. Existing customers still depended on parts and diagnostic access controlled by the company they were suing. The note was outside the materials approved for the AI workflow. Mara had read it separately.
She had intended to reconcile it with the analysis before making her recommendation.
Instead, she had spent the afternoon checking whether the analysis was right.
“Manageable if the replacement comes through,” she said.
“Does it have to come through before they accept?” Dev asked.
The partner looked at the clock. “Ask for a transition provision and they'll reopen the price. We could lose this offer.”
There was no comfortable option hiding behind better wording. More litigation could be financially damaging. Asking for more could leave Bellwether with less. Accepting might leave customers without reliable service.
Mara could feel herself rehearsing an explanation for why the team had missed the deadline. She had not yet missed it. The explanation was already occupying the space where the recommendation should have been.
She opened her morning notes.
Payment. Service continuity. Who can authorize the tradeoff.
The three items had fit on one line at nine that morning. Nothing in them had become more complicated. Acting on them had.
“I'm calling Anika,” she said. “I can give her the financial recommendation. I need her decision on the operating risk.”
Anika answered from a corridor.
“Before you say anything,” she said, “can we please finish this today?”
“I'd like to. The payment and avoided costs support accepting. The agreement also ends the access you need for servicing. We don't have a replacement committed. If you accept, that's a risk the company is choosing.”
“How bad?”
“I can't give you a reliable number from what we have. I can tell you what the agreement takes away.”
Mara explained the provision without reading the summary. She said what she knew, what depended on the unsigned replacement, and what asking for a transition period might cost.
Anika was silent.
Then: “The board authorized me to settle within the financial range. It didn't authorize me to take that operational exposure without bringing it back.”
At 4:56, with Anika's instruction, Mara asked opposing counsel for an extension and a short transition arrangement. The response came before five.
No extension. Their client would consider different terms next week. The current offer would lapse.
Anika chose to let it lapse.
On Monday, the depositions would still happen. Mara would have to explain the additional expense. Nobody had congratulated her.
She saved the call note beside the analysis, including her recommendation and Anika's instruction. Then she deleted the unsent sentence saying the revised terms should be accepted.
The analysis stayed in the file. It was useful work.
It no longer stood in for a decision nobody had made.
Authorial note
Mara had already named what mattered. Under pressure, checking the analysis took the place of bringing that criterion into the recommendation. Dev’s question, her notes and the conversation with Anika helped make the unresolved choice explicit. The costs remain in the ending.
Human Authority examines human capability alongside legitimate standing to act and responsibility for consequences. This story offers a situation to examine. It does not establish a diagnosis or show that an intervention produces better outcomes.
If you are facing a consequential recommendation or conversation, you can explore private work with me. I review applications personally to consider fit. Scope and investment are agreed before work begins.
Fiction and legal context
The people, company, dispute, dialogue, timing and outcome are invented. This is not a disguised client story or a reported case, and it provides no evidence that Human Authority improves outcomes.
The professional backdrop draws on U.S. ABA model guidance. Formal Opinion 512 discusses competent use, confidentiality and review of generative AI output. Model Rule 1.2 addresses client authority over objectives and settlement; Model Rule 2.1 addresses independent judgment and candid advice. These sources do not establish the rules governing any particular jurisdiction. This story is not legal advice or a compliance template. Restricting AI inputs alone does not establish adequate confidentiality safeguards.
Comments