Read time: 5.5 min

One of Well Built’s writers came to me glowing: Deployment Agent once again solved her major head scratcher.

Her love for the tool bubbled over — she’s been using it like craaaazy, all day, every day. 

So… it got me super curious: how helpful is this thing, really?

I asked her to back up and walk me through a use case that really knocked her socks off.

“SURE!” she exclaimed, up for the challenge, “it literally just solved a pickle with generating dynamic promo letters for our midyear review.” 😍

My ears perked up. Adv Comp is my bread and butter (and “how to send out dynamic docs ad hoc” is a solution that lives on Premium 🦄).

I listened intently, asked some questions.

Aaaand the Deployment Agent solution was… okay. As in, mid 😕

Deployment Agent introduced her to a key Adv Comp task. It even solved the surface-level ask. But the full design had downstream impacts and flaws that a seasoned practitioner would know about. I steered her differently. (I did. I’m a human! I can’t stand by and let her implement second best silently!)

I’ve heard dozens of cases like hers, and it HURTS my (very human) heart.

And now, having tested it myself, I’ve got a full take: where Deployment Agent wins, where it falls short, and how you can use it wisely in your day to day 🦉

So, let’s get this straight. It’s not actually an “agent”. It’s a Workday-dedicated AI chatbot, built on Gemini, that launched to Customer Central and impl tenants on March 28th, free to every customer.

It’s backed by the sources you all know and love/hate: Admin Guide, the User Guide, the Training Catalog, accepted Community answers… Allegedly, it’s not trained on customer data, but it’s set to achieve “tenant awareness” later this year so its answers may become more curated and relevant to you 🤞

Recently, Deployment Agent became available in your Sandbox and Preview tenants. This is AWESOME because at the start, you were stuck with that dual-login annoyance since you can’t be in two tenants at the same time without incognito mode 🕶️ Now you get the option to use Deployment Agent in the same place where you’re all likely testing configurations! (Just know that it requires some setup for each tenant.)

Since launch, overall uptake’s been fantastic. Seems like anyone with access is prompting it regularly. 

While success is mixed, the overall sentiment is still positive — instead of digging through Community, submitting an AMS ticket, or asking your coworker, the go-to move is now to prompting Deployment Agent 🤖

(Most of all, people love not going into Community.)

Used right, Deployment Agent's a genuinely good first stop. But it comes with some natural limitations.

And the same goes for these LLMs. The sources that feed Deployment Agent have the same gaps they’ve always had. 

I didn't learn how to build a genuinely creative calc field from Community. I learned it out in the field, using the good ol’ click around and find out method — break it, fix it, get an idea, and repeat the process until it works elegantly. Not a lot of that’s written down anywhere. So Deployment Agent never learned it either.

Those gaps don’t fill in overnight. So, you can expect them to show up (or not show up 😉) in your results. 

On top of this, expect some wrong answers.

You’ve heard about AI “hallucinations”.  The delusional AI chatbot tells you something completely false, but in that exact same “I got this” tone.

This isn’t a bug — it’s how LLMs are designed! They don’t have an “I don’t know” button. They’re built to predict an answer, no matter the cost. The cost when what’s probable and what’s true DON’T overlap? A hallucination. 

Deployment Agent is no exception. Hallucinations have been reported out the wazoo!!! It likes to make up completely fake business objects, fields, tasks, and even calc field functions.

Asking different questions elicits different responses. The same goes for prompting!

For example, I prompted it with — “I'm building out the Assess Skills BP. I want to remove the option to decline rating a skill. How do I do that?”

It correctly replied that it’s not possible to remove the decline rating a skill option. But it went further to say that rating values aren’t configurable. Well, that’s false! — you absolutely can configure (via name overrides) a single set of rating values to apply to all skill assessments across your tenant.

My hunch is that it got stuck on the word “BP”and called out that ratings aren’t configurable to user’s within the BP form experience 🤨

Trying again, I prompted it with — “I want to remove the option to Decline within the Assess Skills form of the Assess Skills BP. Is that possible? Where do I configure that?”

And it gave me the instructions to edit the BP policy and adjust the Deny and Send Back settings. Not what I was looking for ❌

Trying again I prompted — “Within the Assess Skills rating form, is it possible to remove the Decline to Rate option?”

And it finally got to the right answer! ✅

So here’s what I’ve learned: as you prompt, be exact about your request. Include word-for-word titles from Workday that you anticipate could lead to the answer — eliminate keywords that could lead it astray.

Either way…

Every answer Deployment Agent gives you needs a second look. 

ELSE, you could end up with a mess on your hands if you take its recc and it proves incorrect out in the field (like some big 4 consulting firms caught with fake citations in critical, industry leading reports 👀). 

Luckily, a quick visit to verify citations and actually attempting the configuration sets the record straight pretty quickly. It’s a low bar, so make sure you reach it!

But where it’s wrong doesn’t bother me. What's more concerning to me is how folks will adapt when it’s repeatedly right.

Sure, some config is cut and dry. It truly starts and ends with the nuts and bolts. That’s the kind of config you do want to move through quickly, and outsourcing it is exactly the right call.

But just assembling the pieces and hitting the finish line isn’t what made me stand out or earn the nickname “Michael Jordan of Workday consultants” from a former client.

Deployment Agent gives you information, a lead worth vetting. It doesn't give you the decision-quality judgment that goes beyond "it works" into "it works well." ⭐️

That gap — between information and good decisions — is what we call the Workday Gap. This persists, and is exactly the space Well Built exists to close. 

Here's what that looks like in practice: say you want to prevent a certain action on a BP. Deployment Agent will easily walk you through how to build a validation error. But what it won’t recommend (organically) is how you should consider pairing it with help text, so users aren't blindsided the first time they hit the error. 

Best practice? Set your standards with the help text, reinforce it with the error 💪

(Side note: if this recommendation shows up, best believe they’re feeding our newsletters into the model!)

So much of my best config over the years came from playing, testing, and asking, "if this works here, can it also work there?" It comes from putting myself in the end-user’s shoes and asking, “what else would help now and in the long run?”

That's the real danger: these LLMs give the experience of competence when you’re not there yet. The confidence, the speed, the citations stacking up underneath — all of it feels like expertise. But the care that turns "it works" into "it works well" still has to come from you. (Or Well Built, if you hire us 😉)

Deployment Agent can hand you the building blocks. Deployment Agent might get you to the finish line. It just won't get you to good design. Remember, pushing toward an outcome isn't the same as well-built config. That's just go-live energy. And we all know how that goes 😬

In the case of our writer, the answer Deployment Agent unblocked her and gave her a solution that felt like the bees knees compared to what she was facing (manually creating dedicated promo letters). A huge win🏅

But the bigger win would be to move beyond that into well built configuration. And you can only do that by going for the SO WHAT, exercising your creativity, anticipating the downstream impacts, and applying a lot of heart. All the unicorn stuff 🦄 still has to come from you. 

Or someone more seasoned than you. US! 😁

As always, thank you for being a reader.

We’re celebrating you and your pursuit of a Well Built Workday 🥳

Until next time!

the Well Built Team 🌞🏗️