What would you fix first: missed enquiries, failed renewals, or a checkout nobody can find? In Andrew Warner and Jesse Pujji's roundup, ten GitHub projects make those kinds of problems cheaper to test. The episode title promises growth. The more useful takeaway is that each tool needs a specific customer problem, a baseline and someone responsible for the result.
Watch the Full Episode
The directory below links all ten repositories from this episode. Capabilities come from the project documentation. The pilots and measures are suggestions from JQ AI SYSTEMS, not promised outcomes or reported revenue.
Find and Convert Customers
Sellbird
GitHub: ethanplusai/sellbird · Watch the discussion
AI-assisted company research and outreach drafts, with approval before sending by default in the documented setup. First pilot: Compare ten company-level recommendations with a manually researched list. Verify use-case fit and the source behind each recommendation before opening any sending workflow. Measure: Count qualified companies accepted by a reviewer, not addresses collected or emails drafted.
Marketing Skills
GitHub: coreyhaines31/marketingskills · Watch the discussion
Reusable skills for conversion reviews, writing, SEO and analytics. It is a skill library, not an autonomous marketing department. First pilot: Apply one conversion skill to an existing page. Ask for a short brief grounded in approved customer research, then have someone review the copy. Measure: Measure whether more visitors complete the page's main action. Programmatic pages only help when each adds unique value.
Recover and Retain Revenue
Novu
GitHub: novuhq/novu · Watch the discussion
Notification infrastructure for coordinating messages across channels, with an open core and separately licensed enterprise code. First pilot: Prototype one failed-renewal reminder on test invoices. Keep your billing provider responsible for payment state and retries. Compare against its built-in recovery features first. Measure: Count incremental recovered invoices after existing retries, and stop alerts when payment succeeds or the invoice is closed.
Chatwoot
GitHub: chatwoot/chatwoot · Watch the discussion
A shared inbox for support and customer conversations across channels. First pilot: Set up one inbound queue for product or pricing questions. Route purchase-ready enquiries to a named human and keep the support queue intact. Measure: Track qualified conversations handled, first response time and handoffs that were missed.
Mautic
GitHub: mautic/mautic · Watch the discussion
Marketing automation for contact segments, campaigns and follow-ups. First pilot: Move one opted-in onboarding sequence for one customer segment instead of migrating an entire marketing stack. Test rendering, opt-outs and delivery in a test environment. Measure: Track activation and unsubscribes together. Hosting, message delivery and ongoing maintenance still cost money.
Manage Deals and Delivery
Plane
GitHub: makeplane/plane · Watch the discussion
Project tracking for work items, priorities and delivery. The product is Plane, not Plain as one chapter label spells it. First pilot: Give a single team one source of truth for an active project. Have an agent draft status updates against it; keep people responsible for ownership and exceptions. Measure: Compare time spent on status reporting and the number of blocked items that lose their owner.
Papermark
GitHub: papermark/papermark · Watch the discussion
Document links, engagement analytics and data-room features, depending on the edition. First pilot: Share a non-sensitive proposal with a small set of expected recipients. Check link permissions and expiry, then test whether engagement helps prioritize follow-ups. Measure: Confirm interest through a reply or meeting. A view is not buyer intent or proof that a person read the whole document.
Test, Listen and Improve Search
GrowthBook
GitHub: growthbook/growthbook · Watch the discussion
Feature flags and experimentation. Some enterprise code has separate terms. First pilot: Start with one hypothesis, such as whether moving a setup step increases activation. Verify exposure logging and agree on a decision window before launching the test. Measure: Use the chosen outcome and sample requirements. A dashboard cannot rescue a weak hypothesis or missing events.
Formbricks
GitHub: formbricks/formbricks · Watch the discussion
Surveys and customer feedback you can trigger in a workflow. First pilot: Ask one question at a precise moment, such as what blocked a new user after an incomplete setup. Read the responses with the team before automating the analysis. Measure: Look for repeated blockers and test a fix. Survey responses are a signal, not a representative census.
Meilisearch
GitHub: meilisearch/meilisearch · Watch the discussion
A search engine with full-text and hybrid search. The repository distinguishes community and enterprise editions. First pilot: Index a small public catalog or help center first. Save real queries and judge the top results before introducing embeddings or replacing the existing search. Measure: Track zero-result rate and useful top results. If indexing private material, test who can retrieve it before enabling search.
A First Pilot You Can Actually Measure
I'd start with failed renewals only if the business already has enough of them to measure and hasn't solved the problem with existing tooling. For example, Stripe already has automated retries. That changes what a Novu project should promise: a better communication workflow around billing events, not a new payment engine.
- Measure the current workflow. Pick one segment and agree on what counts as a recovered invoice, answered enquiry or successful search.
- Run the smallest test. Use synthetic or authorized data, a test inbox or a separate environment. For reminders, verify duplicate and out-of-order events, and confirm that notices stop after payment.
- Compare with what you already have. Use a baseline or holdout where practical. Count incremental gains, not every sale or recovered payment that followed an automated action.
- Price ongoing ownership. Include hosting, integrations, review time, backups, upgrades and a named owner. Check the license and edition for the features you actually need; public code does not make a hosted service free to operate.
That last step is the agency opportunity Warner and Pujji discuss. You can sell a narrow implementation and dependable support to one type of customer. Replacing several SaaS tools at once buys you more migration work before you know whether the first workflow helped.
CodeShift Companion Library: The Earlier Seven-Repo Episode
The supplied CodeShift resource page covers the earlier seven-repository conversation. It does not list these ten projects, so its media and links are grouped here for further reading. See our separate write-up on that episode for those seven business ideas.
| Companion repository | Topic |
|---|---|
| OpenSEO | Search research and audits |
| Postiz | Social publishing |
| Hermes Agent | Managed agent workflow ideas |
| AI YouTube Shorts Generator | Owned video repurposing |
| VideoLingo | Video localization |
| Ghost | Newsletters and memberships |
| REA | Documenting and rebuilding permitted workflows |
Other links from that resource page: Tribe Digital Co. for a service-business example; FluidVoice, Granola, Grain, Fathom, Otter and Fireflies for capture and meeting notes; and MoreMozi Shorts as an editorial example. Use only recordings and source material you have permission to reuse.
Companion X Posts and Accounts
The CodeShift page links to these authors. Expand an account for its timeline and original posts, or use the direct links if X does not render in your browser. These posts are examples, not independent proof of results.
Video Chapters
| Time | Topic |
|---|---|
| 00:00 | Open-source tools and business growth |
| 00:38 | Sellbird |
| 01:17 | Quality of AI sales leads |
| 02:40 | Manual versus AI prospecting |
| 03:05 | Marketing Skills |
| 04:07 | Making time for marketing |
| 05:23 | SEO and programmatic pages |
| 06:43 | Novu and payment recovery |
| 07:36 | Selling a measurable retention outcome |
| 08:20 | Hyperagent (sponsor) |
| 09:46 | Agent-built pages and videos |
| 10:29 | Chatwoot |
| 12:11 | The open-source agency opportunity |
| 15:06 | Setup and ongoing support |
| 16:37 | Mautic |
| 20:15 | Can customization replace SaaS? |
| 23:57 | Marketing automation services |
| 25:23 | Picking a custom CRM niche |
| 27:10 | Plane and project management |
| 31:33 | Zapier MCP and Papermark |
| 32:00 | Document engagement |
| 34:06 | GrowthBook |
| 35:12 | Better hypotheses for experiments |
| 37:13 | Formbricks |
| 40:29 | Meilisearch |
| 41:40 | The cost of operating search |
| 43:34 | Final thoughts |
Sources and Disclosure
Checked 19 November 2026. The episode and supplied transcript establish the discussion; the official GitHub repositories establish what the tools do. JQ AI SYSTEMS did not deploy the projects, audit their security or verify growth claims. Episode sponsor: Hyperagent. Zapier MCP also appears in the episode. Neither is counted among its ten repositories.