Describe the jobs-to-be-done framing for an AI writing assistant.

Marchvale builds Tonecraft, an AI writing assistant marketing teams use for ad copy, email campaigns, and landing page text. Thessalyn Colfax is the AI PM who owns Tonecraft's roadmap. Silvio Winthrop, Marchvale's CEO, keeps asking for more buttons on the toolbar to match rival copy tools. Verbena Faircastle runs marketing at Amberley Home, a candle and home fragrance brand, and has quietly gone back to drafting Amberley's biggest campaigns by hand.

How to answer this, stage by stage

Nobody is grading whether you can define "jobs to be done." They are grading whether you can name the one real job, in the customer's own words, and the one design decision that makes it checkable instead of a nicer word for a feature list.

Let's learn

Picture two ways to describe the exact same button. One is a list of everything it can do. The other is the one job somebody actually hired it for. Only one of those tells you what to build.

Tonecraft is Marchvale's AI writing assistant. Marketing teams open it to draft ad copy, email campaigns, and landing page text. Before Verbena's team leaned on it, a single big campaign, like Amberley Home's yearly Ember Sale, took her about six hours to draft by hand: pulling from a laminated one page brand voice sheet pinned above her monitor, and a folder of the brand's best performing past emails.

This is the workaround the job screen was built to replace. Not a lack of judgment from Verbena. A lack of anywhere to put it.

Now, with Tonecraft's original nine button toolbar, autocomplete, grammar check, a tone slider, subject line generator, hashtag suggester, hook generator, CTA generator, readability score, and a plagiarism checker, Verbena could get a full first draft in about forty minutes. That felt like a win, for a while.

Two ways to describe the same product. Only one of them tells Tonecraft what it's actually for.

Here's the turn. Forty minutes was never the real problem, and neither were a few rough sentences the tone slider produced. The real problem was what came out the other end: copy that read like every other candle brand's AI written email, warm, a little breathless, nothing like the specific voice Amberley had spent five years building. Verbena started rewriting almost every draft from top to bottom, then started skipping the tool for anything that had to carry the brand's name.

What it costs at its worst: brand sensitive accounts like Amberley are exactly the customers a generic AI writing tool can't easily win, the ones who care most about sounding like themselves and not like a template. If Tonecraft only ever offers a nicer tone slider, those accounts quietly churn to whichever competitor actually asks what job they're hiring for, and Tonecraft is left competing on price with tools that write generic copy for a fraction of the cost.

What Thessalyn built instead: before any draft starts, Tonecraft makes the writer name the job. Match our exact brand voice for this launch. Get a fast draft I'll rewrite myself. Turn one great email into new variants. Picking the brand voice job pulls in Amberley's real style guide and its twelve best performing past campaigns, and every draft has to clear a fact check pass against Amberley's actual product sheet before it can be marked ready, since a model that nails the brand's voice can still invent a claim that isn't true.

The whole fix in one screen. Pick the brand voice job, and Tonecraft pulls in the material that makes that job checkable.

What I would leave alone: for low stakes internal writing, meeting notes, a Slack update, the plain "help me write faster" job is fine exactly as it was, a simple tone slider, no style guide, no fact check. Forcing that work through brand grounding would just slow down something nobody was struggling with.

The lesson: a writing assistant can always do more. That was never the real question. The real question was whether Verbena's actual job, sounding like Amberley without sounding like every other brand's AI written copy, ever had a place to live inside the tool at all.

Now here is the same thing as a story

The short version above is what you'd say out loud in the room. Read this one for what it actually felt like to watch someone quietly stop trusting a tool built to help her.

Verbena Faircastle has run marketing at Amberley Home for five years. She wrote every headline, every subject line, every hero paragraph for the brand's five annual launches herself, and she was good at it. No exclamation points, ever. Never the word "cozy," even for a candle brand, because Amberley's founder banned it in year one for being what every competitor already said. Benefit before feeling, always. She kept all of it on a laminated sheet, and in her head.

Tonecraft arrived two years ago, and for a while it genuinely helped. Late on a Tuesday before a launch, she'd open it, get a rough draft in minutes instead of an hour, and spend the saved time on the parts that mattered most: the one headline the whole email hung on.

Then the habit thinned, in three beats nobody would have called a crisis on its own. First she used Tonecraft for full first drafts. Then only for autocomplete on individual sentences, because the full drafts kept needing a rewrite anyway. Then she stopped opening it at all for anything that had to carry Amberley's name, and kept it only for internal notes, where the voice never mattered.

Nobody decided to stop trusting Tonecraft. It just never had a place to put the one job that mattered most.

There was no single moment. Thessalyn found it during an ordinary quarterly account review, when seat level usage among brand sensitive accounts like Amberley had been quietly falling for weeks, while usage among less picky teams held flat. No ticket. No complaint. Verbena had simply gone back to the laminated sheet, six hours a campaign again, and never once told anyone.

It was never really about the nine buttons on the toolbar. It was about whether any single one of them knew what job it was doing for the person pressing it. A tone slider set to "warm" means the same thing to a brand that wants a fast rough draft and a brand that wants a launch ready send. It can't tell the two jobs apart, because it was never asked to.

Thessalyn remembers the meeting where the original toolbar list got approved, eighteen months earlier. The question on the whiteboard read: what should a good writing assistant do. Someone had already listed what two rival tools shipped, and the room mostly agreed to match it, plus one thing extra. Nobody in that room asked what job any real customer was trying to get done. It made sense at the time. Marchvale had one type of customer then, and getting the basics working was the whole point.

The anchor has to survive its own first mistake. This is what that survival looks like.

What Thessalyn put in its place: before any draft starts, name the job. Match our exact brand voice for this launch, or get a fast draft to rewrite, or turn one great email into new versions. Choosing the brand voice job feeds the model Amberley's real style guide and its twelve strongest past campaigns, and every grounded draft has to clear a fact check pass against the actual product sheet first, the same guardrail that caught a draft claiming a candle was hand poured when it wasn't.

Three weeks after the job screen shipped, the share of Tonecraft drafts Verbena kept without a full rewrite went from 22 percent to 71 percent. The Ember Sale email that used to take her six hours took ninety minutes: one grounded draft, one fact check pass, done.

One nine button toolbar chased "what can a writing tool do." One job screen asked "what job is this for, and can I prove it." The second one is the only version Verbena still opens.

What I'd tell myself, the day that first toolbar list got approved on a whiteboard: the mistake wasn't adding a tone slider. It was never asking, out loud, whose job any of those nine buttons was actually doing.

SPARK, in the words of one candle brand's inboxNot a script for sounding thorough. SPARK is what forces you to name the one job in the customer's own words, and prove the fix survives its own first false claim.

Keep out doesn't mean these jobs are bad. It means bolting them onto this one dilutes what it's actually good at.

The recap, one line per letter: situation is Verbena drafting alone from a laminated sheet, payoff is one habit, name the real job before opening the tool, anchor is the required job screen grounded in real brand material, risk is a vague job producing the same grab bag with a nicer label, and keep out draws the line at any writing job that isn't ad copy, email, or landing pages, no matter how reasonable it sounds on its own.

And if you want to be sure it really works, try it somewhere elseSame five letters, a county permits office instead of a candle brand, and this time the missing job isn't "sound like us." It's "explain exactly why, in words the applicant can actually use."

Almsgate County Permits Office runs Plainscript, an AI writing assistant that drafts applicant facing letters, approvals, denials, requests for more information. Sarafina Warrender owns its roadmap. Almsgate's leadership had built Plainscript the same ungrounded way Marchvale first built Tonecraft: one generic "write a clear, polite letter" setting, tuned for friendliness, with no distinction between a routine approval and a denial that has to cite an exact section of county code.

Mapped onto SPARK: the situation is permit clerks writing every denial letter by hand, pulling the right code citation from a binder, because no tool had ever separated that job from a routine approval. The payoff is the same shared habit, name the real job before drafting. The anchor is the same job screen, run again: is this a routine approval, or a denial that must name the specific code section and the specific fix. Choosing the denial job grounds Plainscript in the actual code text and a set of past letters that survived appeal, not a generic polite tone. Risk runs the same shape: a vague "write a nice letter" job produces a friendly letter with no real citation, which is exactly the kind applicants call the office to argue about. Keep out draws the same line: Plainscript never drafts internal staff memos or press statements, a different job with a different owner entirely.

Swap the trigger and it still runs.
Speed: an interviewer caps you at ninety seconds. Skip straight to the anchor, name the job before the draft, ground the specific job in real material.
Cost: no budget to rebuild the whole toolbar. Ship the job screen alone first, grounding can follow once the highest value job is proven.
The model got better, for real: say a future model writes noticeably better prose on its own. The job screen still matters, because "which job is this draft for" is a question about the customer, not about how good any one model happens to be that quarter.

Where people run it wrong.
They ship one default job so vague it might as well not exist, "write good copy," and call the naming step done.
They let the job screen exist but never actually ground the top job in real material, so picking "brand voice" changes a label, not the output.
They bolt every writing task the company can think of into one job's settings, so the one job that mattered most quietly gets worse to make room for the others.

How to use it live. Before answering a "design an AI writing tool" question cold, ask yourself one thing: what's the one sentence a real person would say about why they opened it today, in their own words, not a feature name. Naming that sentence, not a phrase like "personalized writing assistance," is usually exactly what the question is listening for.

Flashcards (tap any card to flip it)

Check yourself Score: 0 / 0