Building in public: sharing the work behind the result

Building in publicThe practice of sharing selected parts of the work while a company, product or project is being built. In this chapter, the material comes from real decisions, experiments, reversals, evidence and open questions; it doesn't require publishing revenue, personal information or a daily progress report. has somehow become synonymous with posting a Stripe screenshot and waiting for everybody else to feel jealous or inspired. The screenshot proves that somebody paid you, but says nothing about why or how.

Public numbers have something meaningful to say when they explain costs, accountability or how the business works. Buffer, for example, has used revenue and other company figures as part of a much broader transparency practice, to great interest and celebration. (Buffer, 2022)

Building in public can quickly lead to you feeling like a performing monkey, but you don't need to turn your company or your product into a daily show. Public work can be much narrower: you select something real from the week, explain the decision or lesson behind it, and give the people you want to reach a better view of how you understand their problem. They see your expertise through the choices you make.

This chapter helps you decide whether that belongs in the route you chose, which parts of the work deserve to leave your private notes and be broadcast to the world, and what the responses will tell you.

Does public work fit the route you chose?

Your distribution brief should already name the customer, the channel, the action you need them to take, and the time you can give the test. Building in public is one possible action inside that route. It can't repair a channel where your buyers are absent or a product nobody understands, but it can successfully bolster your reputation and thought leadership.

This table compares the situations where public work may productively contribute to getting your product out there:

You can also reject the tactic because you hate it - and that's OK! Founder-led public content is optional; sales conversations, customer examples, research, search, partnerships, communities, a team brand and paid distribution can establish expertise without requiring you to perform a personality that makes you feel uncomfortable online.

Start with the week you had

The source materialThe real work a public piece draws from, including notes, customer questions, decisions, experiments, product changes, results, reversals, specifications and supporting evidence. should exist before the post, and you shouldn't manufacture events in your week to make you feel more interesting. So how do you come up with the source information needed to build this content?

A private weekly record gives you somewhere to collect the things that happened and the work you have completed, including decisions you haven't finished, details that can't be shared and several days that would make appalling content.

Record what you:

• discussed;
• decided;
• locked into a rule, price or specification;
• built or delivered;
• abandoned or reversed;
• still need to answer; and
• heard repeatedly from customers.

You don't have to note all this down manually; that'd be exhausting. You can use Romy's Build in Public Skill, which is a weekly record that reads GitHub commits, Notion and Linear updates, and agent conversations, then sorts them into groups.

One of my own early weeks contained 600 non-merge commits across two repositories, 29 Claude sessions and 119 Codex sessions (excellent work-life balance). Such a mass of activity is useless as a story, which is why the Skill turns the records into Tweets, Reddit posts, blog posts, and other kinds of content.

Find the part somebody else can apply

A shipping update tells people what you did. A stronger piece gives the intended reader something they can recognize or dive into. These are six jobs a public piece can do:

The test is whether the material is true and whether this audience can do something with it. “Day 17: shipped onboarding” is true. If users skipped half the setup, you removed those steps and activation became clearer, the decision underneath the update gives another founder something they can inspect in their own product.

The product can supply the setting and evidence without dominating every piece. It should become the subject when somebody needs to understand or evaluate it, including a launch, demonstration, release explanation, security answer or customer-requested walkthrough. There is no responsible ratio of educational posts to product posts; relevance decides how much product belongs there.

Keep the complete version private

Before drafting, separate what's OK to share, what needs time or permission, and what stays inside the record. The categories below are a working screen, and they aren't legal advice; contracts, regulated industries and different jurisdictions may require stricter rules.

GitLab publishes a handbook built around transparency and still classifies customer contracts, open vulnerabilities, security evidence and sensitive project documents before anything becomes public. The organization is vastly larger than a solo business, so its procedure doesn't transfer, but the underlying decision does. Openness still needs boundaries. (GitLab, accessed September 2026)

Customer experience needs an additional check. General patterns can be taught once identifying details have been removed, while direct quotations, screenshots, company names, recognizable stories and attributed results require explicit permission. In the United States, customer endorsements must reflect the person's honest experience, and payments, incentives or other relationships that could affect how somebody reads the endorsement may require a clear disclosure. (Federal Trade Commission, accessed September 2026)

Personal disclosure is also a choice. You may include your own experience when it changes how somebody understands the decision, although the feed has no claim on your income, family, health or private relationships. A 2024 experiment on influencer recommendations found that credibility judgments changed with the intimacy, tone and subject of personal disclosure. Influencers recommending products aren't solo founders selling their own work, so the study can't provide a founder-content formula. (Leite, 2024)

Use AI to search the week

AI can retrieve the notes, commits, conversations and documents you supply; it can separate decisions from activity, find repeated themes, flag missing evidence, prepare a draft and adapt an approved idea to another format. You can do all this with the above Skill.

You still decide what the work means, which opinion is yours, what can leave the private record and whether the final draft sounds like you. The model can't manufacture customer language, results, motives, vulnerability or a satisfying ending to a week that didn't have one.

One of my weekly records, for example, surfaced a problem in a content workflow: a source could pass through a summary, brief and draft before anybody noticed that its claim had changed. The public subject wasn't the amount of work completed that week. It was the decision to keep the source attached through each step, block important unsupported claims and capture the user's point of view before drafting began. The record showed that those controls had been built and tested against the workflow; it couldn't show that they would catch every source conflict, so that limit would need to travel with the draft.

Let the work set the cadence

There is no universal posting frequency for this chapter. A weekly review may produce three credible subjects, one subject or nothing that should be public yet. Choose one format and a time budget you can sustain, then let the later channel chapters help you test platform-specific frequency.

Silence doesn't mean the business failed that week - like everything else regarding online content, you should only post something when you feel like you have a meaningful moment to share.

Record who responded and what happened next

Views, likes, saves and followers show how far your content traveled. They don't tell you whether it reached your customer.

It's important to record the responses and results from your building in public content so we can take a view on whether or not it's working:

1. Did somebody from the intended audience reply with a relevant question or recognition?
2. Did the piece produce a conversation, introduction, referral or collaboration with a route to the buyer?
3. Did that person visit, subscribe or inspect the evidence?
4. Did they try, request a demonstration, buy or take the customer action named in your distribution brief?
5. How much reach and engagement did the piece receive around those results?

I published a long Reddit post explaining how I approach the route from an idea to first users. My H1 review recorded 8.8k views, 40 comments, two new connections and one consulting inquiry. The view count is the largest figure, and also the most misleading, because eyes on the words don't equate to actions because of them. The engagements, the connections and inquiry are the reason I count the post as commercial evidenceAn action that helps you judge whether public work contributed to the route to a customer, such as a relevant conversation, introduction, inquiry, trial or purchase. Views and likes describe distribution around that evidence; they don't establish demand by themselves.. It's one post, so it can't promise that 8.8k views will produce an inquiry for anybody else.

Your public-work record

For one week, collect the full private record before deciding whether any of it deserves publication. Choose one audience and one job for the content (do you want to reach new people? Ask questions about what you're building to validate your product? Something else?), attach the evidence and its limit, then complete the sharing screenThe decision that classifies source material as share now, delay or ask permission, or keep private. It's an editorial safeguard and doesn't replace legal, contractual, security or regulatory advice.. Download the public-work record for the full worksheet.

The chapter is complete when you have one screened draft and empty response fields ready for anything that follows. Publishing is optional for now.

The next chapter covers launches and Product Hunt, where several distribution actions are concentrated around one release. This record will give that launch something more substantial to draw from than “we're thrilled to announce”.

Prompt: turn real work into one public draft

Use the prompt below with your distribution brief and public-work recordA private record of the week's decisions, work, reversals, questions, customer language and evidence. It supplies possible public material after the founder applies permission and sensitivity checks.. It prepares a draft for your review; it must not publish, schedule or send anything.

I want to turn one piece of real work into a public draft for the people I'm
trying to reach.

My distribution decision
- Customer: [specific customer]
- User, buyer and approver: [same person / different people / I don't know]
- Channel or setting: [where this may be published]
- Customer action I need: [reply / inspect evidence / visit / subscribe / try /
  request a demo / buy / introduce me / another named action]
- Plausible route from this audience to a buyer: [route / I don't know]

My private source record
- Period covered: [dates]
- Discussed: [items]
- Decided: [items]
- Locked into a rule, price or specification: [items]
- Built or delivered: [items]
- Abandoned or reversed: [items]
- Still open: [items]
- Customer language or questions: [exact words, source and permission status]
- Evidence: [links, files, figures and dates]

My boundaries
- Already public: [items]
- Customer or third-party permission I have: [details]
- Confidential, contractual, legal, security or personal exclusions: [items]
- Product claims or promises without approval: [items]
- Material relationships or incentives that may require disclosure: [items]

Your task
1. Extract only the facts, decisions, customer words and evidence I supplied.
2. Keep facts, my interpretation, assumptions and missing information separate.
3. Check whether the intended audience contains buyers or has a plausible route
   to them. Flag the gap if I haven't supplied one.
4. Propose no more than three candidate ideas. Tie each one to a real decision,
   experiment, reversal, result or open question from my record.
5. For each idea, state what the audience could recognize, inspect, decide or do.
6. Classify every source item used as:
   - share now;
   - delay or ask permission; or
   - keep private.
   Give the specific reason.
7. Require explicit permission before using a quotation, screenshot, company
   name, recognizable customer story or attributed result.
8. Reject credentials, access details, open vulnerabilities, NDA material,
   active legal matters and confidential customer information.
9. Recommend one candidate and prepare one channel-neutral draft from it. Keep
   the evidence and its limit visible.
10. Include my product only where it's relevant to the problem, decision or
    evidence.
11. Suggest the smallest next action supported by the material. The draft must
    still work if I remove that action.
12. Give me the response fields to complete if I publish.

Rules
- Don't invent a customer, quotation, result, failure, emotion, opinion,
  disclosure, product behavior or conclusion.
- Don't turn activity volume into achievement without explaining what it
  proves.
- Don't claim that attention, comments or founder approval establish customer
  demand.
- Don't add personal disclosure because it may increase engagement.
- Don't use a universal posting cadence or conversion benchmark.
- Don't publish, schedule or send anything.
- Use American English and ordinary language. Address me as you.

Return exactly these headings
## Facts and sources
## My interpretation
## Assumptions
## Missing information
## Audience and route to a buyer
## Candidate ideas
## Sharing screen
## Recommended idea
## Draft
## Evidence and its limit
## Smallest next action
## Response record

Compare the output with the source record before anything leaves your notes. A model can find the forgotten decision; it can't decide on your behalf that a customer will enjoy becoming content.

Sources

• Buffer. “How Transparency on the Internet Can Support Your Goals.” 4 April 2022.
• Bull, Alice. “How to get your first users (really).” Reddit, 2026. Performance figures from Alice's H1 2026 review.
• Federal Trade Commission. “FTC's Endorsement Guides: What People Are Asking.” Accessed 16 September 2026.
• GitLab. “Transparency by Default.” Accessed 16 September 2026.
• Leite, Francisca Pimenta. “To share or not to share: When is influencer self-disclosure perceived as appropriate?” Journal of Consumer Behaviour, 2024.
• Romy. “Free Agent Skills for the part after shipping.” Accessed 16 September 2026.