Depost AI 2 is live. It is an AI content team for LinkedIn, and a complete rebuild of a product thousands of people already used.
We did not add these features to version 1.0. We stopped working on it and started from an empty folder, because what people kept telling us was not something features could fix.
Here is what they said, what changed, and the four decisions the new product is built on.
The feedback that ended version 1.0
We built the first Depost AI for ourselves. We were not copywriters. We wanted an audience on LinkedIn, and a tool that turned a rough idea into a decent draft felt like the whole problem solved.
Thousands of people used it. Their feedback was consistent, and it was not what we expected.
Nobody said the writing was bad. They said the writing was the easy part.
That is difficult to hear when writing is the thing you built. But it held up every time we checked it. What ate the hour was not the draft. It was everything around the draft:
- Deciding what to write about in the first place
- Making a graphic to go with it
- Picking a time to post
- Being there when the replies came in
- Knowing, a week later, whether any of it had worked
We had automated the twenty minutes and left the other forty exactly where it was. It is the same pattern I wrote about with vibe coding, where the tools make starting easy and leave the hard part untouched.
What people actually asked for
The requests that arrived most often were not about writing at all. Four words came back month after month.
Approvals. Marketing teams needed a review step. Somebody had to sign off before a post went out under the company name.
Workspaces. Agencies were running LinkedIn for a dozen clients out of one login, with no way to keep the brands apart.
Teams. Founders had ghostwriters who needed access that was not the founder’s own password.
Chat. Everyone was tired of picking a tool, filling in a form, and copying the result somewhere else.
None of those are features you can add to a single-user generator. They describe a different product.
Why we rebuilt instead of adding features
Version 1.0 was a set of tools. You opened a generator, filled in a form, got a draft, and took it elsewhere. There was no room in that shape for a second person, never mind a second brand.
Adding to it would have produced a worse version of the thing people were describing. So we rebuilt it.
This is the second product I have chosen to rebuild rather than extend. StartupBase was the first, for the same reason.
Here is what actually changed:
| Depost AI 1.0 | Depost AI 2 | |
|---|---|---|
| How you use it | Pick a generator, fill in a form | Brief a chat in plain language |
| Where a post starts | A blank prompt | Your own analytics |
| Graphics | You made them yourself | Generated in your brand colors |
| Scheduling | A manual queue | A calendar filled from approved drafts |
| Review | None | Approvals, roles, and an audit trail |
| Multiple brands | One account, one voice | A workspace and its own agents per brand |
| Outside the app | Nothing | MCP in Claude, ChatGPT, and Cursor |
| Price | $59 solo | $59 solo, unchanged |
The four decisions behind Depost AI 2
Most of the work went into a handful of calls that were not obvious at the time. These are the ones I would make again.
Publishing goes through LinkedIn’s official API only
The official API is narrower than what a tool can technically do by other means. Working inside it meant saying no to features that would have shipped faster and demoed better.
It is also the entire point. For most of our users, a LinkedIn account is not a marketing channel they could replace next quarter. It is a decade of professional history and the reason clients answer their messages.
So the rule is simple. If something is not possible through the official API, we do not build it.
There is no autopilot
The most requested thing we will never build is a mode where the product runs the account by itself.
Nothing publishes without a human yes. Depost AI sends no automated connection requests, no automated DMs, and no automated comments. Replies and outreach arrive drafted in your voice, and you send them.
That makes the product slower to use than it could be. It is also why people are willing to connect an account they cannot afford to lose.
The price is per account, not per seat
A product built for teams cannot charge per seat. Per-seat pricing charges you for involving the reviewer, the client, and the person who should have seen the draft before it went out. It works against the thing the product is for.
So Depost AI 2 is priced per LinkedIn account:
- Solo, $59 a month, for one account and one seat
- Team, $99 a month for the first account and $59 for each one after, with unlimited seats and workspaces
- Yearly billing is two months free, and every plan includes a seven-day free trial
Those are the same numbers as version 1.0. A full rebuild, no price increase.
MCP ships on every plan
The last thing people told us was that they did not want another tab open. They already spend the day inside Claude or ChatGPT.
So the whole product is reachable over MCP from Claude, ChatGPT, Cursor, or any other MCP client. You create a key scoped to one brand or one workspace, with only the permissions you grant it, and you can revoke it without touching anything else. A research thread becomes a draft in your voice, queued for review, without leaving the conversation.
It is on Solo too. Putting the integration everyone wants behind the expensive plan is a tax on the people who can least afford it.
What an AI content team actually does
The result is closer to a content team than to a writing tool.
It starts from your numbers. Impressions, engagement, and your full follower history sync from the LinkedIn account you connect. Every post is measured against your own average, so what beat your baseline last month is what next week’s ideas get ranked from.
Four agents do the work, and every brand you run gets its own set. One reads the analytics, one finds the angles, one writes the post in your voice, and one designs the graphic in your colors.
You brief it in a chat. Plan my week. Write this in my voice. Interview me about this topic. A full editor sits underneath for when you would rather write it yourself.
The calendar fills from drafts you approved, at times suggested by your own engagement data, with the first comment already written.
Teams get a review step, with roles per brand, threaded feedback, an audit trail, and a separate workspace for every client, so the voice learned for one never leaks into another.
What the rebuild taught us
Three things, and none of them are specific to LinkedIn.
Users report symptoms, not causes. “The writing is fine” was not a compliment. It was a diagnosis, and we heard it as praise for longer than we should have.
A request you get four times is not a feature request. It is the product asking to be built. Approvals and workspaces sat on the roadmap for about a year before we understood they were the roadmap.
A constraint you choose becomes a feature. The official API and the human approval step both started as things we gave up. They are now two of the main reasons people pick us.
Frequently asked questions
What is Depost AI 2?
An AI content team for LinkedIn. You brief it in a chat, and four agents read your analytics, generate ideas, write posts in your voice, and design the graphics. It schedules through LinkedIn’s official API, and every draft waits for a human to approve it. Teams get approvals, roles, and a workspace for each brand.
Why rebuild from scratch instead of refactoring?
Because the problem was the shape of the product, not the quality of the code. Version 1.0 was a set of single-user generators, and approvals, workspaces, and multiple brands cannot be added to that shape. Refactor when the code is bad. Rebuild when the model underneath it is wrong.
Did the price change?
No. Solo is $59 a month for one LinkedIn account. Team is $99 for the first account and $59 for each additional one, with unlimited seats and workspaces. Yearly billing is two months free.
Does Depost AI post or message anyone automatically?
No. Nothing publishes without a human yes, and there are no automated connection requests, DMs, or comments. The product drafts and schedules. You approve and send.
Who is it for?
Founders and consultants who need to be visible on LinkedIn without losing an hour a day, marketing teams who need one voice per brand and a review step, and agencies or ghostwriters running LinkedIn for other people.
Start a free trial
Every plan includes a seven-day free trial, and nothing is held back on the cheaper one.
If you want the full tour, I wrote up the whole release on the Depost AI blog, going through each part of the product and why it exists: Introducing Depost AI 2.
Otherwise, start a trial and brief it yourself.