Innoventer · September 2026 update

Turning Innovent into supervised AI operating loops

Innoventer is our AI operating system for how Innovent gets work done. It connects our products and tools to AI agents that do the work, while people make the calls that need judgment.

Innoventer orchestrates. Products own their domain. Small apps provide specialized interfaces. Humans provide judgment. The test for what we automate: any repeatable process with clear inputs, a definable output, and a way to judge whether the output is good.

The 12 capabilities

They all run on one foundation: agents, a shared board, memory, Slack and per-person logins. The first capability is live and proves the foundation works. The rest are loops we build on top of it.

Working Partly there Next
01

Build software autonomously

Idea → plan → code → test → a PR with proof → deploy. Working today up to the pull request. Merge and deploy are still done by hand.

Working
02

Run product development

Feedback and bugs become specs and approved work.

Partly there
03

Coordinate across the company

Asana, blockers, and a Slack DM when a person's input is needed.

Next
04

Personal command center

What's mine, what's waiting on me, and what can ship.

Partly there
05

Growth and social factories

Research, posts, replies and content calendars.

Next
06

Marketing and business reporting

CAC/LTV, anomalies, executive summaries.

Next
07

Research and news intelligence

Save, research, summarize, brief the team.

Partly there
08

Support and feedback loops

Triage, answer, and turn real problems into engineering tasks.

Next
09

Continuous QA

Agents already test their own work in a real browser.

Partly there
10

Releases and documentation

Every PR already ships with its own docs and walkthrough.

Partly there
11

Work across all our products

PostMoney and VC Match share one factory. RailsGenerator is next.

Partly there
12

Rapidly build internal tools

Small purpose-built apps, wired back to Innoventer.

Next

What we set up, in about a week

One shared system instead of everyone running their own AI tools on their own laptop.

Our own infrastructure

Behind company Google sign-in. Everyone at Innovent logs in as themselves.

An agent per product

PostMoney and VC Match each have one, plus an orchestrator and a dedicated reviewer.

A shared board

Anyone can give an agent a task, watch it work, and pick up or redirect someone else's.

Slack

Message agents directly. #vc-match-eng goes straight to the VC Match agent.

Safe, realistic data

Every task gets its own copy of the app with scrubbed sample data. Agents never touch production or real customer information.

Memory

Agents keep notes and lessons from every task, so the next one goes better.

The software factory loop

What happens after someone writes a task.

  1. Anyone writes a taskOn the shared board or in Slack.
  2. The agent gets its own workspaceAn isolated copy of the app with sample data, so tasks never collide.
  3. It plans, builds and runs the full test suiteThousands of tests, every time.
  4. It uses the app like a customerIn a real browser, on desktop and mobile, and fixes what it finds.
  5. It hands over proofA live preview link, screenshots of every flow, and a plain-English explanation in the pull request.
  6. A person reviews and decidesThen merge and deploy.

How this is different from before

We were each developing alone with our own agents. Now it's one integrated factory.

Before
Now
Each of us worked with our own AI assistant, in silos
One shared factory. Everyone sees the same board, agents and history.
No visibility into anyone else's agent work
We watch each other's work and hand tasks to agents on each other's behalf.
AI wrote code; a person had to set it up, run it and test it
The agent QAs its own work in a real browser before anyone looks.
Reviewing meant pulling the branch and running it yourself
Every PR comes with proof: a live preview, screenshots, test results and the reasoning behind it.
Context lived in someone's head or chat history
Agents remember, and lessons compound across tasks and people.
One task at a time, per person
Many tasks in parallel, each isolated, running while we do other work.

The result: review becomes the job. People spend their time judging finished, proven work instead of producing it.

Proof point

"Chat with your matches" for VC Match

A single task to the VC Match agent produced a complete, production-quality feature. Founders can have private AI conversations about their investor matches: prioritize outreach, compare investors and prepare for meetings, using only the reports they've paid for.

VC Match chat workspace: 'What do you want to decide?' with starter questions to prioritize investors, compare fit and prepare questions
The new match research workspace the agent built, with synthetic data.
1task, no human stepping in
5,329tests passing, zero regressions
52new tests written
5bugs it caught and fixed itself
  • The feature: conversation history, starter questions, cited sources, mobile layout and error recovery.
  • Careful design: founders only see their own conversations and only the reports they're entitled to. If access changes, it fails safely.
  • Self-QA: it used the app like a founder on desktop and mobile, and fixed the bugs it found before opening the PR.
  • Proof for the reviewer: a live preview, an architecture diagram, and a screenshot walkthrough of every founder flow, all with real AI answers.

The pull request as the reviewer sees it

Open full size ↗
Pull request description: design decisions, data and AI boundaries, verification results and screenshots
Scroll inside the frame. Design decisions, safety boundaries and verification, all explained by the agent.

The screenshot walkthrough the agent produced

Open full size ↗
Walkthrough: prioritizing investors, comparing them, preparing for a meeting, on desktop and mobile
Scroll inside the frame. Prioritize investors, compare them and prepare for a meeting, on desktop and mobile.
The same week, agents also built renewal reminders and an email campaign for VC Match (left paused for our approval), a Carta integration and an AI library upgrade for PostMoney, and a PostMoney portfolio profile feature that has already merged.

Where we're going

Build the loops one at a time, each on the same foundation.

Now

Close the engineering loop

  • Review comments go straight back to the agent to address.
  • One-click approve → merge → deploy to staging → the agent verifies the deploy.
  • Add RailsGenerator as the third product.
Next

Connect the company

  • Link agent work to Asana tasks and sprints.
  • When an agent needs a decision, it DMs the right person in Slack and picks up again once they reply.
  • A personal dashboard and a daily Slack digest: what's mine, what's waiting on me, what can ship.
Then

Factories beyond engineering

  • QA: agents regularly use our products like customers and file bugs with steps to reproduce.
  • Support: triage the inbox, answer routine questions, turn real problems into engineering tasks.
  • Reporting: daily and weekly summaries of acquisition, CAC/LTV and anomalies, written as explanations rather than dashboards.
  • Research: save anything interesting and get it researched, summarized and linked to our work.
  • Growth: research topics, draft posts and replies, with approval before anything goes out.
  • Releases: release notes, changelogs and announcements generated from what shipped.

Guardrail: before agents handle customer data or take actions outside the company (sending email, posting), we tighten permissions and sandboxing. Today agents only produce pull requests, and a person always reviews them.