Work with me
Consulting and automation for Open Source adoption, AI-assisted and human-led.
On this page
A question I keep running into: why does a project with active repos and a few hundred stars still have almost no one contributing from outside the team? Usually it isn’t the code. The people who are interested never find a clear way in, and the team doesn’t have the hours to find them and help.
This is how we can work together on that, with the same system that runs Community Signals.
All you bring is your thinking. Your thought dump, meeting notes and call transcripts go into the Brain, and I handle the rest with the system.
You bring
I handle the rest, and you get every week
- Three problems to focus on
- Conversations and newcomers
- Content, no longer the bottleneck
- First issues written for people
- A project in order
Why a prompt on its own isn’t enough
Everyone has the same models now, so the same prompt gives everyone the same post, and people scroll past it. What makes the difference is what you feed in: your team’s raw thoughts, what people are actually saying, and your code.
Could be about any project.
Why your first PR stalls, and the three issues we labelled to fix it
About your project, from what you think and what people ask.
01How it works
You bring your thinking to the Brain. Public discussion and your code come in beside it, AI maps people’s problems to what your code solves, and I decide what goes out.
The work falls into five areas. The Brain is the hub, and the other four draw on it. I handle those with the system. Project and community health come first, with credit and thanks for the work and no points, leaderboards or badges. Later each area can run as its own agent, with a person still deciding.
The Brain
your thought dump, meeting notes and call transcripts, the map and Focus
Community
signals, engagement and onboarding
Content
posts, drafts, ideas and articles
Code
your repos, good first issues, and keeping the project maintained
Governance
roles, credit and thanks, and sustainability
Live now, on my own workBeing built, or set up with you in the pilotA person: you bring the Brain, I decide
From you
Thought dump
notes and mind maps, no need to write well
Meeting notes
or a call transcript
The Brain
the one thing you bring
Public discussion
Bluesky
posts on your topics
Hacker News
stories and comments
Followed feeds
arXiv, GitHub Trending
Community Signals
tone, depth, top news, worth acting on
Your code
Code, docs, issues
your repos, and what they solve today
Code signal
open issues, gaps in the docs, questions people ask
Problem–solution map and Focus
in the Brain: problems linked to your code, three picked each week
I handle the rest
Drafts, checked for tone
your writing samples and rules, on every draft
I decide
accept, edit or decline in the inbox, and check with you
Four areas draw on it
Community
replies from your own account, follow-ups, onboarding
Content
posts and articles, announced from the project’s account
Code
good first issues written for people, a maintenance map
Governance
roles, credit and thanks, the project in order
Measure depth, then go again
Not stars or likes. How deep each contribution goes, read with the Contributor Ladder, shapes next week’s balance.
02Without it, and with it
Outreach done by hand is slow, and most of the week goes on finding things rather than helping anyone. Here is the same week both ways.
The usual wayHuman-led, AI-assisted
Find the conversations
Reply
Follow up in 1 to 3 days
Write an article
Share it everywhere
Write first issues
Change what you track
The usual way~18 h a week
Human-led, AI-assisted~6 h a week
~11 h
saved every week
64%
less time on the same work
+180%
more done per hour
02.1Running on live data
I’m the first user. These numbers come from Community Signals.
1,844
posts read in 11 days, since Sep 4
See the data →60%
wary of AI-written code, against 14% open to it
See the mood →45%
hands-on, the depth where a contributor is worth your time
See the depth →37
posts worth a reply, a talk or an invite this week
See the path →- 01
libvips bans generative AI contributionslibvips on GitHub
- 02
FSFE: can LLM-generated code be licensed as Free Software?Free Software Foundation Europe
- 03
Forgejo keeps AI out of code, docs and conversationForgejo on Codeberg
The system shortlists what’s trending in your space, and a person checks the sources and ranks it. See all
If this already fits your project, we can talk it through.Book a call
03A week, step by step
Step 01 is yours. I run the rest with the system, and this is what it looks like.
Put your thinking in the Brain
This is the part only you can do. Paste your thought dump, meeting notes and call transcripts, or import a mind map. No need to write well. That raw thinking is what makes the output yours, and I handle the rest.
- - people star us, then nothing
- - first PR waits 9 days for review??
- - idea: “first week” issues, 3 max
- - talk idea: AI PRs vs real newcomers
meeting notes · call transcriptmind map · 12 nodes
Listen to the discussion
The system reads public posts on your topics, sorts them by tone and depth, and keeps the ones that matter. This one is real.
“The flood of low effort AI-contributions makes accepting anonymous contributions impractical nowadays. They'd waste hours reviewing things of no value. I know because it happened to several of my moderately successful projects.”
AI contributionsmaintainerwary of AI
One of 1,844 posts since Sep 4.
The Brain maps problems and picks a Focus
What people struggle with, linked to what your project already solves, or could with a small change. Each week Focus ranks the problems by how many people raise them, whether that is growing, the tone, and how close you are to a solution. It picks three, and I keep or change the picks with you.
- 01 Vibe-coded work1,204 posts, 58% wary · suggested: reply
- 02 AI slop in PRs288 posts, 79% wary · suggested: write
- 03 AI code in review259 posts, 44% wary · suggested: open a first issue
Posts and tone from Community Signals. The next drafts and content ideas follow the picks.
Draft, then I decide
For each post worth acting on, AI drafts a reply from your writing samples and tone rules. A check without AI looks for common AI phrasing and em dashes. A draft that fails is rewritten once, and if it still fails, it arrives with the broken rule marked.
Drafts wait in the inbox, and Telegram only sends a link to it. Accept posts the reply on Bluesky, after checking I haven’t already replied. For Hacker News it copies the text and opens the post.
“there's a growing schism in the open source community over llm usage. A LOT of projects are outright banning llm contributions. If volunteers of free projects are rejecting the tech I really doubt that anyone is gonna…”
draft · waiting in the inbox
The bans are mostly about review time, not the tech. A wall of unreviewable PRs costs maintainers hours. If you want to contribute with LLM help, say so in the PR and keep it small. That gets merged. Silent dumps get banned.
Accept posts on Bluesky. Other platforms are copied for me to post.
open with the answer, then the contextfrom my edits and decline reasonsMeasure depth, not likes
The Contributor Ladder reads how deep each interaction goes, so you can see who is ready to climb and where people drop off.
- Surface
- 52%
- Builder
- 45%
- Steward
- 3%
For a client, the same read runs on PRs, issues and reviews.
Being built: two accounts, first issues, the project in order
Two accounts, each with its own voice
An evangelist account is a person’s own voice: replies, commentary, community engagement and follow-ups. The main account is the project’s official voice: releases, updates, notes, the digest and article announcements. Each has its own tone rules and writing samples, and every draft says which account it goes out from.
evangelist account · reply
I had the same problem on my own project. A short review checklist in CONTRIBUTING.md helped newcomers more than anything else I tried.
main account · release note
New release. CONTRIBUTING.md now has a review checklist, and three issues are labelled for first-time contributors.
First issues, written for people
Good first issues suggested from your repo: open issues, gaps in the docs, questions people ask in public. Each one is written for someone new to the project, and favours work that rewards understanding it, so the way in suits people rather than generated pull requests. A person approves each one before it is opened.
Add a review checklist to CONTRIBUTING.md
Beside it, a maintenance map: pull requests waiting on review, stale issues, questions that keep coming back, docs out of step with the code.
Keep the project in order
Roles suggested from the Contributor Ladder over time, like triager, reviewer, docs lead or maintainer, each invitation a draft a person sends. Credit and thanks for every kind of work: code, docs, reviews, triage, answers and talks. And checks on the project’s basics, maintainer load, and work that depends on one person.
LICENSECONTRIBUTINGCODE_OF_CONDUCTSECURITYGOVERNANCEMAINTAINERS
Solid: present and current. Dashed: missing or out of date.
draft · thank you
Thank you for answering the setup questions in Discussions. It saved the maintainers a lot of back and forth.
draft · role invitation
You’ve been reviewing docs changes for a while now. Would you like to be listed as a docs reviewer?
03.1From conversation to contributor
A like or a follow ends there. A conversation can go further, if someone follows it up at the right time and with the right next step. This is the path, what I do at each step, what the system does, and what gets measured.
01
A conversation
I reply to someone stuck on a problem your project touches. Helpful, not a pitch.
From my own week: I asked a developer whether they use specs or prompts. They explained their setup and shared the tool they use.
02
A follow-up, within 1 to 3 days
I share one useful thing: a guide, a discussion, or where to ask.
03
An invitation
I point them to a first issue that fits what they told me.
04
A first contribution
Your team reviews it quickly, and I thank them in public.
05
They come back
We give them context and a bigger piece of work.
What gets ignored, and what gets a reply
In evangelism, the listener should never feel like a transaction. It’s a big turn off, and people can tell. Here is the same path both ways.
03.2The contributor funnel
Stars, likes and follows are arbitrary metrics. They show how a project looks from the outside, not its impact, how deep someone goes, or how much a contributor cares. Teams are looking for impact now, so this follows people from a conversation all the way to contributing, and every step is measured.
What looks good from outside
Stars, likes, follows
Easy to count, easy to inflate, and silent on who stays.
What shows impact
Depth
Replies that get an answer, second contributions, time to a first review, and new Builders and Stewards.
Conversations read
1,844 so far, from Community Signals
Worth acting on
37 this week, sorted by AI
Replies and articles
posted once a person accepts
Visits to your repo
GitHub traffic, by referring site
First issues picked up
labelled issues
Pull requests
merged, and how fast
Builders and Stewards
read with the Contributor Ladder
03.3Publish once, everywhere
Every article goes up on your own site first. Copies go out from there, each linking back to the original, so search engines credit you.
Your article
on your site, the original
dev.to
imported as a draft, linking back
Hashnode
linking back to the original
Bluesky
a short post with the link
Mastodon
a short post with the link
drafted for you to post
X
drafted for you to post
03.4You steer what it hears
Change the phrases yourself at any time, or approve the ones the system suggests from the posts that worked. For research like Community Signals, the phrases stay fixed so the trends compare, and any change is dated.
See the phrases, and this week’s suggestions
open source communityopen source maintainersmaintainer burnoutnew contributorscall for speakerscall for proposalslooking for contributorslooking to contribute+ add phrase
- submit your talk10 matches in 7 days
04Is this for your project?
Early AI and devtool Open Source teams
Active repos, people starring the work, and few outside contributors yet.
Funded commercial Open Source
Seed to Series A, wanting adoption and contributors rather than more ad spend with no clear return.
04.1How do we start?
I work as an evangelist. I help first, before selling anything, because adoption in Open Source has to happen naturally. People stay with a project that helped them.
The usual way
Sell first
Launch posts, ads and cold pitches, then hope people stay.
The evangelist way
Help first
Be useful in the conversations people are already having, and invite them in once they’ve seen what you know.
A call
where your project is, and where it’s stuck
4-week pilot
- Week 1Read your code, docs and community
- Week 2Set up signals and the Brain
- Week 3First replies, content and issues
- Week 4Report on what moved
Monthly
I run the agreed mix, and we revisit the balance
04.2Where does it run?
On your own server, so your data stays with you. Or I run it for you during the pilot.
See what runs on your server
Listens to
- Bluesky
- Hacker News
- GitHub
- arXiv and feeds
Your server
Dashboard
the inbox where you accept drafts
Listener
reads new posts every 12 hours
One database file
yours, export it any time
Your AI provider
the one you pay for, with a monthly cap
One install per team. Nothing is shared with other clients.
I approve
Accept, edit or decline
nothing goes out until I accept it
Then posts to
- Bluesky
- Mastodon
- dev.to
- Hashnode
Built on Open Source
- Docker
- SQLite
- Astro
Tracking on X is an add-on for narrow topics, since its API is billed per request.
Where it is today
- Live nowListening and sortingBluesky, Hacker News and research feeds, sorted into posts worth acting on
- Live nowThe inboxreply drafts in a self-hosted dashboard, for my own use
- Live nowTone ruleswriting samples, banned phrasing, a check on every draft
- Being builtThe Brainthought dump, meeting notes, the map, Focus and the two accounts
- Being builtCode and governancegood first issues, maintenance, roles and credit
- Being builtContent and communityideas and articles, follow-ups and onboarding
04.3What do you get every week?
You bring the Brain. Each week you get five things, each one measured.
Three problems to focus on
From the Brain each week: the problems people raise most, ranked, each with a suggested action. I keep or change the picks with you, and the drafts follow them.
measured: how each picked problem moves, week on week
Conversations and newcomers
Discussions worth joining, with draft replies from your own account, followed up until people find a way in, and calls for speakers in your space.
measured: replies that get an answer, first and second contributions, talks submitted
Content, no longer the bottleneck
Ideas and articles from your notes and the week’s picks, in your voice, announced from the project’s account on dev.to, Hashnode, Bluesky and Mastodon.
measured: posts published, reads, visits sent to your repo
First issues written for people
Good first issues from your repo, each saying where to start and what a finished contribution looks like, plus a map of what needs maintaining.
measured: issues picked up, days to a first review
A project in order
Roles suggested as people grow, credit and thanks for every kind of work, and CONTRIBUTING, SECURITY and the rest kept current.
measured: roles filled, work that depends on one person
At the end of the pilot, you get a written report with the numbers for each part we ran.
05Book a call
Tell me about your project and where it’s stuck, and we’ll see if this fits.
Book a callor email ask@joshuapoddoku.com
I’ll always ask before I share or write about your project. If you’d like to follow along first, there’s also Patreon.
I’d welcome your thoughts, questions or critique. And if you know a project that could use this, I’d be grateful for a referral.
Thank you to the maintainers and developers whose public posts shape Community Signals, and thank you for reading.
Book a call