Finally, Marketing That Makes Sense to Engineers

I launched Quiver today.

It is an agentic developer-marketing system, but the shortest explanation is the one on the homepage:

Finally, marketing that makes sense to engineers.

I have been writing code since 2006, and I have spent much of my career marketing developer products. That combination changes how I work.

When I run into the same problem twice, I want to build the thing that removes it. When important information lives in five places, I want a source of truth. When a draft becomes production, I want a state change—not a document someone renamed “FINAL-v7.” When a campaign ships, I want to know what happened and carry that evidence into the next decision.

That is normal engineering practice.

Marketing software somehow decided none of it applies.

The problem was never generating more marketing

AI made it remarkably easy to produce marketing output.

Give a model a prompt and it will give you positioning, campaign ideas, landing-page copy, email sequences, social posts, competitive analysis, and a strategy deck before lunch.

Then the conversation ends.

The positioning stays in one chat. The customer research lives in a document. The campaign plan is in another tool. The approved copy gets pasted into a CMS, where it loses the reasoning and evidence behind it. Results appear in a dashboard weeks later, disconnected from the work that produced them. The next AI session starts by asking you to explain the company again.

We solved generation and ignored the system around it.

The result is more output, not necessarily better marketing.

I felt this problem at full force while running GTM for Tabstack. I was doing the work of a team: positioning, developer experience, competitive research, content, launches, outreach, analytics, and the systems connecting all of it. I built agents to gather information and draft work, but I kept the judgment and action for myself.

That approach worked. In my first 90 days, signups increased 71%, activation increased 29%, API requests grew almost 6x, and website traffic increased 129%.

Quiver was part of the operating system behind that work. It began as the GTM command center I needed for myself: one place where the product context, research, sessions, campaigns, content, and results could stay connected.

I open-sourced it in April. Then I kept using it.

And as I used it, the actual product became obvious.

Quiver was not another AI writing tool. It was the architecture missing from agent-powered marketing.

I would never build software this way

Imagine running a software system the way most teams run marketing.

The source of truth is whatever someone remembers from the last meeting. Every agent gets a slightly different configuration. There is no version history. Development and production are the same environment. Deployments have no state. Logs are stored somewhere else and rarely make it back to the people writing the next release. When something works, the learning disappears into a slide deck.

No engineer would accept that system.

Yet technical founders are regularly handed marketing as a collection of vibes and disconnected tactics.

Post more. Start a newsletter. Run a campaign. Make a content calendar. Try a different headline. Feed the website into an AI tool and ask it to sound less generic.

The problem is not that these activities never work. The problem is that there is no architecture connecting them.

I built Quiver around the primitives engineers already trust.

A source of truth, not another blank prompt

Every useful marketing decision depends on context: what the product actually does, who it serves, how those people describe the problem, what the company believes, what evidence exists, which claims are safe to make, and which hypotheses are still unproven.

That context should not have to be reconstructed every time someone opens a chat window.

Quiver keeps positioning, ICP, messaging, customer language, proof points, brand voice, and active hypotheses in one approved product context. Every session starts from that state.

Agents can propose changes when new evidence appears, but they do not silently rewrite what the company believes. A person reviews what becomes true.

The point is not to remove judgment. It is to stop wasting judgment on reconstructing the same context over and over.

Version history for the work and the thinking behind it

Marketing changes because the evidence changes. Positioning evolves. Customer language gets sharper. A campaign disproves an assumption. A new product capability changes the story.

That does not mean the old state should disappear.

Quiver versions the product context and the artifacts created from it. You can see what changed, preserve the lineage, and restore an earlier version when needed.

That matters even more when agents are involved. If an agent creates or changes something important, I want to know which context it used and what happened afterward. “The AI wrote it” is not an operating model.

Explicit states between an idea and production

A draft is not approved. Approved is not live. Live is not archived.

Quiver gives artifacts explicit production states:

Draft → Review → Approved → Live → Archived

That sounds simple because it is simple. It is also one of the pieces I care about most.

AI tools often collapse generation and execution into the same moment. The model produces something, and the surrounding product encourages you to ship it immediately. That may be convenient, but it removes the boundary where human judgment belongs.

Quiver keeps that boundary visible.

Agents can gather, analyze, recommend, and create. Humans decide what is true, what deserves to ship, and what the results mean.

Campaigns as the spine, not a folder name

A campaign should be more than a label on a group of assets.

In Quiver, campaigns connect the plan, agent sessions, customer research, artifacts, content, tasks, and performance. You can trace the work from the original decision through what shipped and into the result.

That relationship is what makes the system useful later.

When I open a new Strategy, Create, Feedback, Analyze, or Optimize session, the work does not begin in isolation. The agent can operate from the approved context and the relevant history around the initiative.

The conversation becomes part of the system instead of disappearing when I close the tab.

Content with an API, because content is infrastructure

I care an unreasonable amount about Quiver’s Content API.

Most content tools treat the publishing destination as the source of truth. The content ends up trapped inside a website, and everything around it—versions, campaign relationships, source material, distribution history, repurposing, and performance—gets scattered elsewhere.

Quiver keeps the markdown, production state, SEO and social metadata, tags, authorship, distribution records, repurposing lineage, and metric history together.

Then the public Content API serves approved work as structured JSON.

Your website controls the presentation. Quiver remains the editorial source of truth.

That means the same approved content can power a Next.js site, documentation, a resource library, or another interface without turning copy-and-paste into the integration layer.

The Content API is fricken gold.

Customer research that can actually change the next decision

Customer calls, surveys, support tickets, reviews, social posts, and field notes contain the language and evidence marketing needs. Most teams summarize them once, put the summary in a folder, and move on.

Quiver turns that research into reusable system inputs: themes, sentiment, Voice of Customer quotes, product signals, and evidence for or against active hypotheses.

The important quotes stay available to the people and agents creating the next campaign. Evidence does not become a positioning change just because a model found it interesting; it becomes a proposal for a person to review.

Again, the system helps with the work. It does not take the decision.

Observability and a feedback loop that actually closes

Marketing teams are good at reporting results and surprisingly bad at preserving what those results should change.

A campaign ends. Someone presents the numbers. The deck gets filed away. Three months later, another person makes the same decision without the learning.

Quiver connects performance to the work that produced it. When something goes live, the system creates a reminder to measure it. You can log quantitative results and qualitative observations, synthesize what worked, and review proposed updates to the product context.

Nothing silently trains on your company. Nothing rewrites the source of truth behind your back.

The system improves because the evidence, decisions, shipped work, and results stay connected—and because people decide which learning carries forward.

That is the loop:

1. Start from approved context.
2. Use research and evidence to make a decision.
3. Create and review the work.
4. Ship it through an explicit state change.
5. Measure what happened.
6. Carry the approved learning into the next cycle.

AI is useful at every stage. Human judgment still owns the system.

Agents should operate the system, not become another place work disappears

Quiver includes purpose-built sessions for Strategy, Create, Feedback, Analyze, and Optimize. It also exposes the system through MCP.

That means an external agent can work with Quiver’s context, campaigns, artifacts, content, customer research, tasks, and performance through structured tools.

This distinction matters to me.

A tool with an MCP does not have to be another destination I remember to visit. It becomes a capability my agent can use. But Quiver still holds the operating state, permissions, and history around the work.

In hosted Quiver, teams get a ready-to-connect MCP endpoint with OAuth or scoped tokens. The open-source edition includes the MCP server code, but you deploy and expose it yourself.

Either way, the agent operates through the system instead of creating another disconnected pile of output.

Quiver does not do your marketing for you

I want to say this plainly because AI products tend to overpromise here.

Quiver will not supply your judgment. It will not understand customers you never talk to. It will not manufacture a point of view. It will not make a weak product interesting or turn unsupported claims into truth.

You still have to do the marketing.

Quiver gives that work architecture.

It helps people and agents share the same approved context, keep evidence connected, move work through real states, preserve the history, and learn from what happened.

The system makes me faster. It does not replace me.

That is the whole reason it works.

Open source for builders. Hosted for teams.

Quiver is available in two forms.

The open-source edition is MIT licensed and free to self-host. It is a strong fit for technical founders and startups comfortable running the application, database, authentication, updates, and MCP server themselves.

Hosted Quiver is the expanding team product. It gives you a shared, always-on workspace, simple team invitations, managed infrastructure and authentication, a ready-to-connect MCP endpoint, and hosted-only features including built-in tasks.

Founder is $49 per month for up to three seats. Team is $99 per month for unlimited seats. Both include the current hosted feature set and differ by seats, not a maze of feature gates.

You bring your own model provider account, so you choose the model for each job and keep control of the provider relationship and its data policy.

Who I built this for

Quiver is for technical founders and developer-tool teams who know marketing matters but cannot stand the way most marketing software thinks.

It is for the person using agents heavily and watching useful context disappear across chat histories.

It is for the developer marketer handling strategy, research, launches, content, outreach, and measurement with more work than one person can carry manually.

It is for the team that wants AI leverage without giving an agent permission to quietly redefine the company or ship whatever it generates.

And honestly, I built it for myself.

I have spent years turning repeated GTM problems into systems. Quiver is the place those systems finally meet: context, research, campaigns, content, tasks, publishing, and performance, with people and agents working through the same operating model.

The problem was never that I needed more AI output.

I needed the work to stay connected.

Now it does.

Start a 14-day hosted trial or self-host the open-source edition.

Bring your own model account. Keep the judgment. Run developer marketing like a system.