We take the drudgery. You get back to your craft.

Three things, mostly. It analyzes information you already collect but never look at. It catches the follow-up that slips: customers, projects, people, commitments nobody circled back to. And it drafts the communication you write over and over, from the same emails and reports to the marketing that keeps sliding to next week. What that adds up to is hours back—hours your team spends on the work that actually moves the business.

In practice: a year of sales or service records turned into a plain-English summary of what's actually happening. A meeting recording turned into action items with names and dates attached, so nothing dies in somebody's notes. A month of social posts and a newsletter drafted from work you've already done, in your voice, for you to edit rather than start from scratch. The follow-up emails nobody has time to send, drafted and ready for review. The same customer question answered for the fortieth time without you typing it again.

What it doesn't do well is judgment, relationships, and anything where being wrong is expensive. Those stay with you.

The point isn't really the hours—it's what they were being spent on. Most owners, and most of their best people, lose a chunk of every week to work that doesn't need them: retyping, reformatting, chasing, reminding. Hand that off and the same team spends its week on the work that's worth their time, which is usually also the work they'd rather be doing.

No. Size isn't the variable that matters. What matters is whether you do some things over and over, and whether any of it lives somewhere other than your head: an inbox, a spreadsheet, whatever software you already run on. A three-person shop with a busy inbox has more to work with than a twenty-person company where everything is tribal knowledge. If your work leaves a trail, there's something here for you.

It works at both ends of the range. When one person is the bottleneck for everything, handing off even two recurring tasks changes the week measurably. When a small team is running a company much larger than its headcount, the same handoff multiplies across everyone who touches the work. We've mapped opportunities for trades, clinics, retailers, nonprofits, and professional services, and the patterns repeat across all of them.

There is an honest floor, though it isn't about size. If nothing about your operation is captured anywhere, no software, no spreadsheets, no email trail, then there's nothing for AI to work from yet. That's a systems problem, not an AI problem, and we'd tell you so rather than sell you something that won't stick. It's rarer than you'd think. Most businesses are sitting on more usable information than they realize.

No, and for most businesses that's usually not the real question. Teams aren't overstaffed, they're overloaded. What AI takes over is the work that eats hours without needing anyone's judgment, and what comes back is capacity. It isn't about replacing people. It's about unlocking them to do the work you hired them for in the first place.

Worth being straight about: AI is a tool, and tools get used differently depending on who's holding them. Some large companies are using it to reduce headcount. We're not going to pretend that isn't happening.

Most businesses aren't in that position, though. They're not carrying extra people. They're carrying people stretched across three jobs each, with a backlog nobody ever gets to. If you're watching costs, the saving usually isn't a position you eliminate. It's the position you don't have to add for another year, and the revenue you stop leaving on the table once follow-up actually happens.

If you're an employee reading this rather than an owner: the parts of a job that change are the parts most people were glad to be rid of. What's left is judgment, relationships, and decisions. That part isn't going anywhere.

No. If you can use a computer for ordinary business work, email, a spreadsheet, a web browser, you're technical enough. These workspaces run on plain English. You upload the documents and spreadsheets you already have, sometimes a screenshot, and you ask for what you want in the same words you'd use with a new hire.

The part that actually matters isn't technical skill, it's willingness. Someone on your team has to be open to doing a familiar task a new way for a few weeks while it becomes habit. That's the real requirement, and it's the reason most AI attempts stall out.

Where you land on that changes what we build, not whether we build it. Some clients want to learn the concepts as we go, so they can run and improve things without calling anyone. Others would rather have it built and handed over with clear documentation and steps to follow. Both are fine. We just need to know which one you want.

Nothing to start. The first call, the discovery session, and the written opportunity report are free and yours to keep whether or not you hire us. Paid work comes in two pieces. Goldwork Foundations, training plus a fully provisioned AI workspace, is $1,249. Most implementation projects land between $2,500 and $7,500 depending on how many workflows we build, and a single straightforward one runs less. Fixed project fees, never hourly.

The range is wide because the work is scoped to you. The opportunity report gives you a written list of what's worth building, and you decide how much of it to pursue. One simple workflow puts you at the low end or under it. Half a dozen involved ones puts you at the top.

Beyond that, the only ongoing cost is software. Your workspace runs on one of the major AI platforms, typically Claude, ChatGPT, or Gemini, at roughly $20 to $30 per person per month. You need fewer seats than you'd expect. Only the people actually running the workflows need one, which is sometimes just you. Some projects add one more tool, a site builder for landing pages, say, at another $10 to $20 a month.

That's the entire picture. We don't sell software, we don't take referral fees, and we don't bill hourly. Subscriptions at this level tend to justify themselves quickly. If a workflow gives one person back two hours a week, the math stops being close.

Some clients keep us on monthly after the build, as the person they hand new AI ideas to as they come up. That's optional, and it comes later. Nobody starts there.

Yes, though not on the first day. Building a workflow takes a session or two of your time: showing us how the work actually gets done, deciding what good output looks like, checking it against real examples. After that it runs. Before you start, do the arithmetic on whatever task you have in mind. How often you do it, times how long it takes. That number is what's on the table.

The second half of that question is the one worth taking seriously. Most tools become another thing to manage because they're another system: another login, another set of settings, someone who has to own it. These workflows aren't that. When something needs to change, it's usually a matter of editing plain-language instructions, swapping in an updated document, or adjusting what the output should look like. Closer to revising a note than administering software.

We document how to do it, and we'll teach whoever's running it to make those changes themselves. When something needs to go further, hand it back to us. What you shouldn't end up with is one more dashboard nobody opens.

The time works in your favor over the long run. Setup is paid once. The hours come back every week after that, and they compound as more of the work moves over.

Less than people expect, and more than the hype admits. Set up properly, it can know your market, your customers, your reviews, and how you sound. What it can't do is own the outcome. It has nothing riding on the decision and no reputation exposed if it's wrong. Give it a clear task and a good example and it will be right, consistently and checkably. The limit shows up where being right is a judgment call rather than a match to a spec.

Take writing, the place most businesses feel the gain first. It'll critique a draft honestly, tell you the opening is slow, catch a tone that's off, and get you from blank page to something workable faster than anything else you've tried. What it won't do is care. It doesn't know that this is the wrong week to send this, or that this particular customer needs handling, or that the whole idea is one you'll regret by Thursday. Work that nobody's judgment touched reads that way, and people notice.

Two other honest limits. It can only work from what exists somewhere: written down, exported, uploaded, or online. The knowledge that lives only in your head has to get out of your head first, which is part of what we do together. And there's work where it should never be the last stop. Final numbers, contract terms, anything with a compliance or licensing dimension, anything you'd defend to a regulator. It can prepare all of that. It shouldn't be what decides it.

And sometimes the honest answer is that a workflow isn't worth building. Some of what surfaces during discovery, we recommend against, because it's fragile, or it's rare enough that setup costs more than it returns, or a person is simply better at it. You'll see those in the opportunity report alongside the ones we think are worth doing.

Sometimes it will. The useful question isn't how to guarantee it never happens, it's whether you'd catch it when it does. We build workflows so the output can be checked against what it came from: a draft you scan before it goes out, a table you can trace back to the rows it was built from. An error you can see is a small problem. An error that hides is the expensive one.

The failure people worry about is hallucination, where a model states something confidently that isn't true. It happens most when a model is answering from general knowledge rather than from something in front of it. That's a large part of why we provision a workspace with your actual documents, data, and examples: work grounded in your own material is far more reliable than work invented from scratch. It doesn't make the risk zero, and we won't tell you it does.

How much checking a workflow needs depends on what happens if it's wrong. Drafting with a person approving before anything goes out is the usual starting point, and for anything customer-facing or consequential it's where we'd leave it. But not everything needs a person in the middle. An internal summary that goes to three people on Monday morning can run end to end, because the cost of an occasional mistake is that somebody notices and mentions it. We make that call per workflow, with you, based on the actual stakes rather than a blanket rule.

When something does go wrong, it should only go wrong once. A bad output usually means an instruction was ambiguous or an example was missing, both of which are fixable at the source. That's the difference between a workflow you maintain and one you babysit: the mistakes make it better instead of accumulating.

Less than you'd guess, and it's front-loaded. The discovery session runs about an hour. Foundations is a session or two. During a build, expect a few hours spread across a few weeks, mostly spent explaining how things work now and reviewing what comes back. Once a workflow is running, your time in it drops close to zero.

Your time is the real constraint on this work, not ours. Most of what makes a workflow good lives in someone's head, and the only way it gets in is by someone explaining it. That's the hour that can't be skipped or delegated, and it's why we ask for a specific person rather than a committee.

How that's spaced is up to you. Concentrated sessions move fast. An hour or two a week works too, it just runs longer on the calendar. What we'd ask for is enough momentum that the project doesn't stall between sessions, because a half-built workflow is worth nothing.

Five steps, and the first three cost you nothing.

1. A 15-minute call. What you do, where the time goes, whether there's anything here worth pursuing.
2. A discovery session. The slow pass through how the work actually gets done, week to week.
3. A written opportunity report. What's worth building, what isn't, and what each one is worth in hours.
4. Foundations. Training, plus your AI workspace provisioned with your real documents and data.
5. Build. The workflows themselves, documented, with check-ins at 30 and 90 days.

Most engagements run about six weeks, though the range is genuinely a few weeks to a couple of months, depending on the pace you set.

You decide how far down the list to go. The opportunity report is a menu, not a package. Some clients build everything on it, some pick one workflow, some take the report and sit with it for a quarter. All of those are reasonable, and the report is yours regardless.

What you end up with is the working thing itself, plus documentation on how to run it and change it. Then we check in at 30 and 90 days: is it still being used, is it still doing what it should, has anything drifted. We're not finished when the build is delivered. We're finished when the work is actually running the way it was supposed to.