Prosperaize

We gave up a three-month build and got two-hour proposals, 15-minute onboarding and a quieter inbox.

We run Level 2 on ourselves: one AI workspace, connected to our knowledge, tools and design systems, used by every role in the company. It came after a custom build we stopped. On the work we measured, it is about ten times faster: proposals from days to two hours, questions to the founding partner from tens a day to one or two, content from hours to minutes.

~200 hours to set up Level 2 across the company
Client
Prosperaize, our own company
Who uses it
Practice, engineering, growth, HR, project and design leads
Level
Level 2, after a Level 4 build that stopped
Built with
Claude Team, connected to Confluence, Jira, Figma, Google Drive and Semrush
Time so far
About 200 hours: 100 from the founding partner, 100 from the rest of the team
Status
Live, and growing with our knowledge and processes
Published
8 November 2026

What changed

Our own estimates, from the founding partner and the team, November 2026.

2h

to a client proposal or deck, on our own design systems. It used to take days.

Founding partner, Nov 2026

15min

of someone’s time to onboard a new person.

Founding partner, Nov 2026

1–2

questions a day still reach the founding partner, down from tens.

Founding partner, Nov 2026

~200

hours to set up across every department, against two engineers for three months on the build it replaced.

Our own estimate

01 · Where we started

“We tried doing marketing at Level 4, with a custom build we would productionize. Level 2 turned out much more comprehensive, and better.”
Dušan Stamenković, founding partner

We started where people would expect an AI firm to start: a custom marketing system we would build and run ourselves. Two engineers worked on it full time for three months. It planned and wrote marketing content step by step: a campaign, a shortlist of ideas, a plan a person edits, then the draft, for blog and LinkedIn first. We were to be its first client, and then sell it to marketing agencies and content teams.

At the end of those three months we saw it wasn’t the right solution. We had built it around types of posts. Every new feature took more engineering time, then more maintenance, and connecting it to our own tools meant building and running those connectors (Model Context Protocol servers) ourselves. Meanwhile the subscription we already paid for could do more than the build, once it was connected to where our work lives.

“We kept building for the next type of post. Every feature meant more engineering, then more to maintain. The workspace took on new work the day someone wrote down how we do it.”
Dušan Stamenković, founding partner

~200

hours to set it up so far: about 100 from the founding partner, 100 from the rest of the team.

Our own estimate, November 2026
ConfluenceJiraFigmaGoogle DriveSemrushOne AI workspaceClaude Team · rules, access,design systemsPractice leadsEngineeringGrowthHRProject managersDesign

02 · The rules

The work wasn’t the assistant. It was the rules around it.

Where things live, who sees what, and how pages connect decide the quality of every answer.

01Every space has an audienceCompany, how we work, methodology, clients, growth and products each have a space. Client spaces are open only to the account partner and the team.
02Pages link, they don’t repeatEvery page sits in the hierarchy, and a process links page to page instead of copying content.
03Access follows the personEach person connects with their own account, so the assistant sees only what that person may see. Everyone’s permissions are checked the way an enterprise would check them.
04One source for the brandDecks, pages and documents start from our design systems, not from a blank page.
05What stays outAnything prospects or clients see. Our AI Investment Diagnostic answers them in our name, so it is a separate Level 4 build: case study 05.

03 · The readout

The build we stopped, against the workspace we kept.

Both judged on what they do for the people who use them every day.

Level 4 against Level 2 · our own company

Same company, same goals

The readout: the build we stopped, against the workspace we kept.

Judged onLevel 4: a custom marketing buildLevel 2: one configured workspace
EffortTwo engineers, full time, three monthsAbout 200 hours across the team
Who it servesOur marketing, then agencies as customersEvery role in the company
What it coversBlog and LinkedIn posts, from idea to draftProposals, decks, content, onboarding, questions
A new kind of workMore engineering, then more to maintainWrite down how we do it
Connecting our toolsConnectors we would build and runConnectors that already exist
Keeping it currentA product to maintainGrows with the knowledge base
VerdictStopped: built around types of postsLive, across the company

The build stopped after three months. The workspace keeps growing with our knowledge and processes.

Our own judgment and figures, November 2026

04 · How we use it

Every role, every day.

Practice, engineering, growth, HR, project management and design.

01

A client proposal or deck in about two hours, from days, built on our own design systems.

02

Posts, pages and briefs in minutes instead of hours, from what we already know and have written down.

03

A new person onboarded with 15 minutes of someone’s time.

04

Questions that used to reach the founding partner answered from the knowledge base: one or two a day now, down from tens.

05

A conference presentation built by voice, hands-free, on a four-hour drive, and on brand.

06

This website’s redesign, designed in the same workspace.

05 · What it shows

Level 2 is real work. Here, it paid more than the build.

When the answer is a configured subscription, we tell clients so. We took the same advice ourselves, after paying for the build first.

Weeks, not daysConnecting sources, access, permissions and rules took about 200 hours, and the work keeps going.
The rules make the answersWhere things live, who sees what, and how pages connect decide what the assistant can do.
Every role, not one teamThe same workspace serves practice, engineering, growth, HR, project management and design.
Check the subscription before the buildWe paid three months of engineering to learn it. A Find engagement asks that question first.

If this sounds like your company

When the seats are paid for, and the knowledge is everywhere else.

You pay for AI seats nobody has connectedThe assistant can’t see your documents, tickets or designs, so the answers stay generic.
A custom build is on the tableCheck first what the subscription you already have can do once it’s connected.
Questions keep reaching the same peopleThe answers exist. Nobody can find them without asking.
Decks and proposals take daysEach one starts from a blank page instead of your own material and brand.

Find out what the seats you already pay for should do.

Schedule a call with a partnerFree until 11 December