Not a Workflow Company, Part 2: Why Cleaner Software and Full Automation Are Both Wrong Turns

Part 1 in our “not a workflow” series drew a simple line: Envoy is not a workflow company, and our execution layer, Ellie, is not another tool built to make reps manage the same work on a cleaner screen. She sources capacity, runs the outreach, negotiates inside your guardrails, verifies the carrier, and gives the rep room to stay in the conversation that still needs a human. 

Envoy is built around execution, not workflow theater.

In conversations about this, though, I usually find myself pushing back in two directions. One group hears “not a workflow company” and immediately turns it into a software cleanup discussion: fewer tabs, fewer logins, one prettier place to manage the same work. Another wants to make the answer bigger: If Ellie can do that much, why not take people out of the process altogether?

Both reactions come from the same frustration. In my career, I used to pay some of the most expensive people in my building to spend their afternoons sending the same emails to the same carriers, chasing the same replies, and checking the same details. Cleaner software starts to sound useful when you live that every day, and full automation starts to sound inevitable.

Neither path, though, solves the actual problem. Cleaner workflow leaves the work intact, while full automation strips out the judgment that makes great reps valuable in the first place. The real opportunity sits between those extremes: removing repeatable execution without removing the people who still create value. 

That’s the idea behind Ellie, and it’s where Part 2 begins.

A Tidier Set of Screens Is Still a Dozen Screens

Cleaner software assumes the floor is buried in complicated processes. It isn’t. The work is scattered across too many places that don’t connect. A single load lives in a dozen systems: the TMS, two or three load boards, the carrier portals, email, chat, and the phone numbers a rep keeps in his own head and his own spreadsheet. He pulls the rate in one place, finds a truck in another, logs into a portal to book it, then digs the rate con out of his inbox, retyping the same load number from screen to screen all afternoon. 

That isn’t a complicated process. It’s a simple job smeared across nine tools. A workflow product does little more than organize that smear, rather than ending it. Hand a rep a tidier set of tabs, and he’s still the glue holding a dozen of them together by hand.

One More Login Doesn’t Help

Drop another workflow tool on top of all that, and you’ve handed the floor one more place to log in.

Ellie works the other direction. 

She runs inside the tabs the rep already has open and handles the portals straight through the browser, the way a coworker would, with nothing new to learn and nothing to rip out. DAT’s 2026 outlook lands in the same place. It tells brokers the way through is cutting cost per load through automation, tighter carrier vetting, and dynamic bidding at scale. 

None of that gets easier when you add a screen. The freight has to move without a rep touching each one.

“Full Automation” Is a Fantasy

There’s a darker reason the phrase “workflow company” bugs me. “Workflow company” sits one short step from “full automation,” and that’s the pitch I trust least in this whole category. 

Automate every step, pull the human out, tell a brokerage it can run on half the floor. I spent years in carrier sales. That pitch is make-believe. The wrong carrier on the wrong load, a rate that should never have left the building, a compliance check nobody ran: in freight, those have always cost real money. 

What’s more, double brokering is as bad as it’s been in years, and a black box that doesn’t actually understand freight can’t reliably tell a real carrier from someone re-brokering your load for a cut. 

The fact that some of these mistakes now cost you in court should have you paying very close attention.

After Montgomery, “The Software Did It” Isn’t a Defense

In May, the Supreme Court held, 9-0, that a broker can be sued under state law for negligently choosing an unsafe carrier. The case is Montgomery v. Caribe Transport II, and we broke down what it means in our Montgomery piece. 

Every carrier you book now carries legal weight, and that’s not a call I’d hand to a black box. A load ends in a wreck, a plaintiff’s attorney wants to know how you vetted the carrier that caused it, and “the software did it” won’t hold. “Here’s the MC and DOT we pulled, the safety screen that cleared, the guardrail that fired, and the rep who signed off” will. 

That record is the observability I argued for in Part 1, kept in the memory and policy layer beneath Ellie. After Montgomery, it’s the line between a defense and a settlement.

The Rep Was Never the Problem

Both wrong turns make the same mistake. One buries the rep under a cleaner pile of screens. The other tries to delete the rep altogether. Get a good rep on a call he’s had a hundred times with a carrier he trusts, and you’ll see the real asset: the one who answers on the first ring, the read on whether a rate is real or a stall, the sense that a load’s about to roll before the tracking says so. 

None of that lives in a workflow, and none of it survives full automation. Trust is the real barrier, not capability. PwC found 28% of leaders rank it among the top obstacles to getting value from AI, and it climbs as the stakes climb. Put a person in the seat, and the floor starts trusting the agent, because they watch her do exactly what they’d have done, only faster.

Call it an execution layer. Call it the operating model that the next cycle rewards. Just don’t call it a workflow.

Book a demo with Ellie to see how she works.