The Shortcut That Wasn’t

Elena Cross didn’t do small talk before bad news, which Tom had always respected about her, right up until the moment her bad news landed on him.

“Two weeks,” she said, over a video call that had started three minutes late and would end, Tom already suspected, in the same place most calls like this ended. “The regulatory filing references the updated eligibility flag. If the API doesn’t expose it, we don’t have a filing. We have a gap, and it’s my department’s name on the finding.”

Tom had the Core API’s documentation open in another window, the same page he’d been staring at for two days. The flag Elena needed wasn’t there. It had never been there. Someone would have to add it.

“I hear the deadline,” Tom said. “I don’t control who can build it.”

“I thought that was your team.”

“It’s not our API. We consume it. The team that owns it is Wesley’s team.”

“So talk to Wesley.”

Tom had. That was the part he hadn’t said out loud yet, because saying it out loud made it real, and once it was real, somebody was going to have to explain how they’d gotten here.

Nate Bishop was already in the room, half-listening, half-typing, the way he did when he was building an argument in his head before anyone had asked him for one. He looked up.

“We asked already,” Nate said. “They said they have no capacity.”

Elena’s expression didn’t change, which was somehow worse than if it had. “None?”

“None,” Tom said. “Not ‘low.’ Not ‘in six weeks.’ None.”

“Then we build it ourselves,” Nate said, the way he said most things, like the idea had arrived a half-second before the sentence did. “Small service, wraps the one field we need, exposes it to the filing system. In and out. We don’t touch their layer, we don’t wait on their backlog, we ship on our own clock.”

Elena looked, for the first time on the call, like she’d heard an answer instead of an excuse. “Can you do that in two weeks?”

Tom felt the shape of the next hour arrive before it happened.

The answer to why Wesley’s team had zero capacity, not low, not deprioritized, zero, took less time to find than Tom expected. It took exactly one phone call, to the same person who’d once picked up his phone on a Tuesday afternoon and, in four minutes, handed Tom the fix to an outage nobody else could find.

“Arvind’s team asked for two of my backend engineers,” Wesley said. He didn’t sound defensive about it, just precise, the same precision Tom remembered from the first time they’d ever spoken. “Then it became three. Something about centralized logging and tracing, standing it up under the Operational Excellence banner. I’m not against better observability, Tom. I’ve been asking for it myself for years. I just don’t have anyone left to fix bugs in the API everyone still depends on while we build the dashboard that’s supposed to tell us it’s broken.”

“Arvind Rao?”

“Senior Director, reports up to Vikram. Cross-cutting mandate, whatever that means this quarter. He’s not wrong that we need real visibility into this stack. I’d have told him that myself, if he’d asked me before he asked my engineers.”

Tom knew the name the way most of the building knew it, from a slide with a single word on it in a font size clearly meant to be noticed. Visibility. He’d sat in the room himself, months back, and watched nobody answer a basic question about logging while that word sat there behind Arvind’s head, unremarked on, undefined. He hadn’t expected to run into the same word again, from the other direction, as the reason a team had no capacity left to fix the thing that was actually broken right now.

“Did anyone flag what’s still depending on your team while your people are off building someone else’s dashboard?”

“I flagged it,” Wesley said. “Told him we’ve got at least three initiatives hanging off this API right now, one of them’s a compliance filing. He said, more or less, that visibility work benefits everyone eventually. I don’t disagree with the eventually. I disagree with doing it by hollowing out the team that keeps the lights on today.”

Tom sat with that for a second. Somewhere upstream of a two-week deadline that hadn’t existed a month ago, a man with a mandate about visibility had made a call about where his own visibility didn’t quite reach, the actual load his reassignments were leaving behind. The consequence of that call hadn’t landed on Arvind. It had landed on Wesley, then on Nate, then, in about ninety seconds, it was going to land on Tom.

“So there’s no version of your team touching this in two weeks,” Tom said.

“There’s no version of my team touching this in two months,” Wesley said. “I’m not being difficult. I’m being honest. I don’t have the people.”

Nate’s plan had one property Tom couldn’t argue with: the code itself really was small. A thin service, one endpoint, wrapping a single field. Tom’s two strongest engineers could write it, test it, and have it running in a sandbox by Thursday. He said as much, out loud, in a room with Nate and Grant, and watched Nate’s shoulders drop with relief before Tom finished the sentence.

“Then we’re fine,” Nate said.

“We’re fine on the part that takes minutes,” Tom said. “We haven’t talked about the part that takes months.”

Grant, who had been quiet in the corner the way Grant was quiet in most rooms where a fight hadn’t started yet, looked up. “What part takes months?”

“Getting it into an actual environment anyone’s allowed to depend on. This company doesn’t let code near production because someone finished writing it. It has to exist first. Officially. In the CMDB. With an Application Unique ID, a security review, an architecture review, a PROD-readiness sign-off. None of that cares how fast the code got written.”

“How long does that take?” Grant asked.

Tom actually had a number, which turned out to be almost worse than not having one. “Around two months, last time I went through anything close to this. But that was before the AI governance rewrite last year, and that rewrite didn’t just touch AI projects. It tightened the whole intake review, top to bottom, because half the point was catching AI-assisted builds that looked finished before anyone had actually reviewed them. Doesn’t matter that this service has nothing to do with AI. Same gate, same reviewers, same new checklist.”

“So more than two months.”

“Some amount more. I don’t know how much. And we just moved onto a new version of ServiceNow, which is what actually runs the CMDB. I haven’t touched an intake request since the migration. I don’t know yet what’s the same and what isn’t.”

He heard himself say it and understood, for the first time, that he was about to spend the next few days finding out something he’d assumed, right up until this conversation, that he already knew.

It took three days, and at the end of them, Tom had learned exactly one useful fact: the number in his head was wrong, and nobody could yet tell him the right one.

Not because they were hiding it. Because between the AI governance rewrite and the ServiceNow migration, half the people who used to be able to answer this question were now answering it two different ways depending on which week they’d last actually touched the process. He searched the internal wiki, three different ones, because the company had somehow accumulated three, and found exactly one page that addressed intake timelines directly. It was dated fourteen months earlier, described a version of ServiceNow that no longer existed, and referenced a review board that had been reorganized twice since. He closed the tab within a minute of opening it, more unsettled by finding something than he’d have been finding nothing. A wrong answer, confidently written down, was worse than an honest blank.

He asked in the architecture channel and got two answers that contradicted each other from two people who both sounded certain, one of them still describing the pre-migration workflow without seeming to know it.

He escalated the question to his own leadership chain, the way you’re supposed to, and got a response four days later that read, in its entirety: checking on this, will circle back.

Nobody circled back.

What finally worked wasn’t a channel or a ticket. It was a name, passed to him sideways by someone on Wesley’s team who’d absorbed part of the reassignment: Dana Okafor, over in governance, who’d apparently run this exact process by hand for six years and had never once been asked to write any of it down, because everyone who needed to know it already knew to ask her directly.

Tom messaged her expecting a delay. She answered in eleven minutes and offered him thirty on her calendar the same afternoon.

“You’re the first person who’s asked me this since the ServiceNow cutover,” she said, once they were on the call. “Usually people find out the hard way, halfway through a project, when their date already exists and I’m the one who has to tell them it was never real. This month it’s worse. Even the people who think they remember the process are working off a version that doesn’t exist anymore.”

“I noticed.”

“Then let’s walk the actual one.”

Dana didn’t need slides. She had the whole thing in her head, the way people do when they’ve explained it enough times that the explaining itself has worn a groove, and lately, relearned enough of it that the groove had shifted under her too.

Every new application, she told him, goes through what her team called an AIA, an Application Intake Assessment, before it gets anywhere near the CMDB. Security questions first. Data classification. Hosting model. Vendor exposure, if any. Depending on the answers, it routes to Architecture, InfoSec, or both, for a real review, not a rubber stamp, a review that checks whether the thing being proposed actually fits the patterns the rest of the environment depends on.

“Two months, start to finish, used to be a fair estimate,” Dana said. “That’s clearly the number you had in your head, and a year ago you’d have been close. Two things changed since. First, the AI governance rewrite. It doesn’t just apply to AI projects, it applies to the intake process itself, so now every application gets an extra layer of review checking whether AI was involved anywhere in how it was built, tested, or is going to run, before anyone even gets to the parts that were always there. Doesn’t matter that your service has nothing to do with AI. You still answer the questions, and someone still has to sign off that the answers are true.”

“And the second thing?”

“We’re six weeks into a new ServiceNow instance. New workflow engine, new approval routing, half of it not fully mapped to how the old one worked yet. Requests are sitting in queues that used to auto-route and now don’t, because nobody’s finished rebuilding every rule. I’m not telling you that to complain. I’m telling you because it means even I don’t have a clean number right now. I have a range.”

“What’s the range?”

“Four months, if nothing bounces back and the new instance behaves itself. Six, more realistically, especially anything that touches a queue we haven’t fully verified yet. I’d plan around six and be pleasantly surprised.”

Tom already knew, before she finished the sentence, exactly what this meant for the two weeks sitting on Elena’s calendar. It didn’t mean the deadline was aggressive. It meant the deadline had never touched reality at any point in its short life, and the version of reality Tom himself had been carrying around wasn’t much better.

“Why isn’t any of this written down anywhere current?” he asked.

Dana was quiet for a second. “Because it’s always been faster to just ask me than to write it and keep it updated. That was fine when the process didn’t change much and there were three of these a quarter. It’s not fine during a year when the process itself changed twice. Nobody’s made it anyone’s job to keep a living version of this, so it stays in my head, and every time something upstream shifts, everyone who thought they knew the old version finds out the hard way, same as you just did.”

Tom brought Dana’s range into the next status call the way he’d once brought someone else’s email into another one, not as an objection, but as a document with a name and a number attached, because arguments without those tend to evaporate in rooms like this one.

“Four to six months,” he said. “Best case. And I’ll own this part, that’s not the number I walked in with either. I was carrying a two-month estimate from the last time I actually went through this myself, and it isn’t close anymore. Two things changed since, the AI governance rewrite and the ServiceNow migration, and neither one made it onto anyone’s radar as something that would touch a project like this. That’s not a negotiating position. That’s what the process actually requires right now, and it requires it whether we like it or not.”

Elena’s face didn’t do the thing faces do when they’re hearing something they already suspected. It did the other thing, the one where a person realizes a problem is bigger and older than the meeting they walked into. “Then what was the two-week option ever going to be?”

“Nothing that could survive contact with this company’s own rules,” Tom said. “Nate’s design was sound. The build itself really was small. But ’small to code’ and ’fast to exist’ stopped being the same sentence around here a long time ago, and apparently they got further apart in the last year than even I’d noticed.”

Grant, to his credit, didn’t try to defend the timeline. He also didn’t say what Tom half-hoped he’d say, which was Arvind’s name, out loud, attached to the actual chain of cause and effect: a cross-cutting mandate about visibility, pursued by pulling capacity from the one team whose visibility actually mattered right now, which had pushed Nate toward a shortcut, which had run straight into a process nobody had bothered to document, current, because it had never needed to be anyone’s problem before.

“So what do we actually do,” Grant said instead, “with two weeks on the calendar and no version of this that fits inside it?”

It was Elena, not Grant, who answered. “We stop asking IT to solve a scheduling problem I created by not asking the right question sooner. I’ll take the real timeline to the filing team. If they need something sooner, they can tell me what actually has to move, the filing date or the scope, because it isn’t going to be this process.”

“And Wesley’s team?” Tom asked.

“I’ll raise it,” Elena said, which wasn’t a promise, and everyone in the room understood it wasn’t a promise. It was the sound of a decision moving one level further up the chain than any of them could see from where they sat.

The filing date moved. Not because anyone particularly wanted to explain that upward, but because there was no version of the alternative that didn’t involve someone eventually asking why a two-week promise had been made about a process that took four to six months, and nobody wanted to be the name attached to that answer.

Arvind’s reassignment of Wesley’s engineers didn’t get reversed. Tom hadn’t expected it to. Decisions made under a cross-cutting mandate rarely get undone by the team quietly absorbing the cost, no matter how clean the cause-and-effect chain looks on paper once someone finally draws it out. What changed was smaller, and Tom suspected it mattered more in the long run than a single reversal would have.

Dana got asked, for the first time in six years, to actually write the process down, and to keep it current, a living version instead of a page that would go stale the next time a platform migrated or a policy got rewritten out from under it. Not because leadership had a sudden appetite for documentation, but because three separate people, in the space of one bad week, forwarded Elena’s after-action summary to their own leadership with some version of the same line attached: this should have been visible before it cost us a filing date. Visibility, it turned out, moved faster once it had a receipt, which was a sentence Tom found he could not think without also thinking of a slide with that same word on it, behind a man who still hadn’t answered for what it actually meant.

Tom never found out whether Arvind ever heard the full version of how his own initiative, the one with Visibility in oversized type on a slide, had cost the building visibility somewhere else entirely, through Wesley’s team, through Nate’s shortcut, through Elena’s deadline, all the way down to a governance analyst’s calendar. He suspected not. Titles like Arvind’s had a way of moving on before the bill came due, and Tom had learned not to spend much time waiting around for that particular bill to arrive.

He didn’t spend much time being angry about it, either. He’d learned, a while back now, that the anger was optional and the fix wasn’t. The fix, this time, was a process finally getting written down, kept current, and a filing date that reflected something true instead of something convenient. That was smaller than justice. It was still worth having.

People like to say a deadline failed because of unknowns nobody could have predicted. That’s rarely what actually happens.

Most of the time, the unknowns aren’t unknown at all. Somebody in the building already knows exactly how long a process takes, exactly what a decision upstream will cost the team standing downstream of it, exactly why a two-week promise was never going to survive its first contact with reality. The information isn’t missing. It’s just invisible, sitting in one person’s head instead of somewhere anyone else could go find it, and invisible information gets treated, by everyone who doesn’t have it, as if it doesn’t exist.

There’s a sharper version of that same trap, and Tom walked straight into it before he ever got to Dana. He wasn’t operating on no information. He was operating on old information, a number he’d earned the hard way once and then quietly assumed would still be true, because carrying it forward had never felt like a decision at all. Processes change. Platforms get upgraded. Policy gets rewritten in response to a completely different problem and lands, sideways, on every project in the building, whether or not it has anything to do with the reason the policy exists. Expertise has a shelf life, and the expensive kind of not knowing isn’t the honest blank. It’s the confident, outdated answer nobody thought to double-check.

That’s the trap underneath the trap. Leadership didn’t fail here only because the world handed them a surprise. They failed because they committed to a date without asking whether anyone in the building, Tom included, was working from something current, and it turned out the real answer belonged to someone who updates it by hand, every day, because nobody had ever made it her job to keep it somewhere the rest of the building could see.

The second failure sits underneath the first, quieter and easier to miss. A decision about where capacity goes, made by someone with the authority to make it and a mandate that sounds like exactly what the org needs on a slide, doesn’t stay contained to the initiative it was made for. It travels. It shows up on someone else’s impossible deadline two teams later, and by the time it lands, the person who caused it is nowhere near the room where the cost finally comes due, often still talking about visibility, in a different meeting, to a different room, with the same word still undefined.

Neither of those problems gets fixed by working faster. Code has never been the bottleneck here, not really, not since the tools got good enough to write the easy part in an afternoon. The bottleneck was always visibility, the real kind, not the kind on a slide: whether the process that governs your promise is written down somewhere a person can find it before the promise gets made, kept current enough to still be true, and true regardless of how certain anyone in the room, even the most experienced person in it, already feels that they know the answer.