How to Post on Social Media from Claude

Learn how to post on social media from Claude using Buffer.

Last week, 72 people at Buffer stopped their regular work and spent the week building.

It was our third company Build Week, and our most impressive, with 112 projects completed.

This Build Week was different for us. In previous years, we’d pair into teams to work on a few pre-selected projects. This time around, we decided to have each person use AI to assist them with building and have everyone direct their own project. This allowed everyone to follow their own learning path while still leaning on experts across the company.

By Friday: 57 new internal tools live on our own domain, from 41 different people. 1,083 pull requests merged across 26 repositories. 15 changes shipped into our admin product. Six people merged their first-ever pull request. And many more ‘side quests’ were completed.

It was a monumental week for Buffer, and this post is a recap of everything we built. We’ll be sharing a bit more about Build Week in the coming weeks — reach out to us on social if you have a specific question.

1. Work someone was repeating every week

• Customer Advocacy Scheduling by Hannah. Replaces our Advocacy coverage spreadsheet and an old Slack command, pulling approved time off so the grid shows who's genuinely online. Its best feature is one a spreadsheet could never have: everyone can read the whole thing in their own time zone.
• Buffer Love, in the Wild by Sabreen. A wall of real things customers have said about Buffer — paste a link, drop in a screenshot, and it fills in the quote, the summary, and the features they praised.
• Content Calendar by Kirsti. One hub for everything the Content team publishes, bringing blogs, newsletters, and social together with month, pipeline, and agenda views. Card tints deepen and fade with a piece's status, and you can push social posts straight into the Buffer queue.
• Ensemble by Juliet. The control center for Customer Office Hours, replacing a Notion database, several Slack workflows, and a lot of manual reminding. It also produces a report we didn't have before.
• Carta benchmark automation by Suzanne. Our quarterly compensation benchmark pull used to mean downloading 34 separate CSV files and pasting them into a spreadsheet. Claude now does it.
• Weekend Update by Octa. Friday-to-Sunday activity, so advocates start the week with context, posted automatically on Sunday evening with the data refreshed just beforehand.
• Automatic incident surveys by Raf. When an incident resolves, everyone who took part now gets a short survey, because in practice, only the incident lead writes the post-mortem. Rather than build a new app, he extended Tally, our internal survey tool, so surveys can now be created from a webhook for anything.
• Public incident write-up skill by Raf. Takes the internal post-mortem and drafts the public status page version against our template, validated with our brand voice skill. The reviewer gets a file to adjust before publishing.
• Culture digest bot by Lexi. Scans all of our culture channels monthly and writes one snackable highlight per active channel without naming names, flagging new channels and nudging the quiet ones. It also builds a live channel guide for each new hire when their onboarding checklist reaches that step.
• Pulse by Kateryna. A lightweight Slack check-in that asks each person two rotating questions every few weeks, so we get a continuous read on how the team is doing between the big surveys. It syncs the roster and time off, so it never messages someone while they're away.
• Prorate by Nate. Works out the refund owed on an annual plan from the price and the dates. It had been living as a Claude artifact that a few advocates passed around, and it's now a tool anyone can reach.
• A guided voice interview by Dave. An interactive tool that learns how you write and speak, then generates a voice skill with far deeper context than a style guide. The framing came from a comment by Jess: we don't just want our writing to sound less like AI, we want it to sound more like us.
• Ash, an Ashby screening agent by Simon. Hiring managers configure what "good" looks like for a role, and Ash reads every application and scores candidates on completeness, role fit, and cultural alignment, then commits the result back to Ashby in one click. It tracks how often the human agrees with it over time. Currently in security review, so it runs locally.
• Design at Social pipeline by Alicja. A complete flow for designers to make carousels about how we build Buffer: a Claude project holding the agreed themes, a design system with slide variants, and two templates. The one honest gap is that moving assets between the two Claude surfaces is still copy and paste.
• Social asset export skill by Alicja. Converts an HTML carousel export into JPGs that are ready to post.
• Social Studio by Andy. Create on-brand Buffer social images from templates, then export a PNG.

2. Docs someone had to remember to update

• Tool Access by Mick. A directory of the 133 tools we have accounts for, answering whether a teammate has an account, what level of access they have, and what needs to happen if they leave. Where a tool's API allows it, the page reads live data every time you open it, so the answer can't go stale.
• Team directory by Hamish. A brand new teammate directory, and the important part is one line: no more data pulled from Notion. Our own app is the source of truth now, which means you update your own details in it.
• Teammates Dashboard by Kasia. People Ops views of teammate data, starting with time off per person and per team.
• Team Calendar by Jenna. One place to see what's happening across Buffer and what's coming, with category filters, month, week, and list views, and a toggle for everyone's Bufferversaries. The bigger idea is that each event eventually carries its own context — the project, the recording, the post.
• Advocacy by Åsa. Answers in one place what Customer Advocacy is trying to do and how it's actually going, with strategy as the home page, so a newcomer meets it before any numbers. Each key metric carries two comparisons and a 12-period trend line, and the current inbox reads Help Scout live.
• Cool School by Sabreen. An interactive campus for Buffer's Team of Creators, built to replace the static resources hub we keep in Notion. Find your angle, your platform, and your rhythm at your own pace.
• HC Changelog MCP by Kelly. A connector anyone can add in Settings, showing every change made to the Help Center over any seven-day period.
• HC Review Queue and HC Audit by Kelly. A queue of suggested Help Center edits driven by product changes, and an audit that surfaces the five oldest articles for accuracy. Both are paused: fixing a bug surfaced real questions about security and write access, and she'd rather wait for the answer than ship past it.
• Buffer API What's New by Justina. Makes our API changelog readable, with a playground so you can run each example against your own account and a share button that drafts a post. It's built with AI and uses AI inside as a feature.
• Recommended content from HQ by Rathes. Moved our recommended-content prototype off a Notion database and a GitHub Actions sync, onto real data straight from HQ.
• Openly by Hailley. Reads what we're already talking about internally, groups it into idea cards, and checks each one against everything we've published on this blog, so a good idea doesn't slip by.

3. Answers that took too long to find

• Ask Design by Kate. One hub for every brand resource, because our rules, guidelines, and assets currently live across Figma, Notion, and nested Dropbox links. It leads with search, so "what hex code is our brand green?" gets an answer.
• How We Marketed It by Maggie. Look up any Buffer feature and see how it's been marketed across product, email, and social.
• Onboarding request triage by Ben. Hand it a support conversation, and it separates every request in the thread, classifies each one, says whether we can accommodate it, and drafts a reply. It points to our live handbook rather than containing a copy, so there's nothing to update when a process changes.
• Billing Context by Nate. Adds a note to billing conversations laying out what actually happened to someone's subscription, reconstructed from Stripe events and Mixpanel. It also suggests whether we should approve a refund, based on our own Help Center guides and leaning toward generosity.
• High-value call brief by Daniel S. Give it an organization and get back a one-page brief for the two minutes before a call: plan and channels, what they've contacted us about, what's failing, what's open. His write-up is the most useful one on the board, and most of it is about being wrong — he built six rows and deleted all six.
• Customer insights library by Mike E. A database autopopulated by an agent reviewing NPS responses, support tickets, research interviews, and in-app feedback, organized so an agent can read it back. Install the plugin, and you can ask why customers switch to Buffer, or how they feel about our pricing.
• Advocacy UI extension by Mau. A browser extension that rebuilds the interfaces advocates live in, starting with Mixpanel — platform, caption, media specs, error labels, and IDs all at a glance, instead of one click per event. He used it himself all week while troubleshooting for customers.
• Popcorn Token Atlas by Daniel P. Describe what you need a color to do — "helper text under a field" — and get the web, iOS, and Android token side by side. The bet is that if a token records what it's for, a model has something to reason about instead of guessing from a hex value.
• Sourcing for hiring by Steph. An attempt to invert the search: rather than waiting for applications, find people in the technical space who already have a public voice. Out of a pool of 3,000 people, she has five names worth looking into, which is itself a useful finding.
• Overlap by Hailley. Finds the best time to meet across our time zones using declared working hours and time off rather than reading anyone's calendar. It defaults to protecting whoever's worst off in a slot rather than optimizing the average.

4. Things we couldn't see about the product

• Customer Themes by Darcy. Took 11,428 customer conversations from one month and turned them into 24 themes, saying for each whether the fix is a better help article, a product change, or an expectation we set wrong. It found that 135 of 136 questions about failed posts were answered with zero negative feedback — and 98 of those customers escalated to a human anyway.
• Checkup by Dace. A consolidated view of errors across Buffer, showing which areas cause the most and what's safe to filter out as noise. It also tracks how long we take to go from an error happening to acting on it.
• API health dashboard by Joe B. Four views of our public API: adoption and growth, which features are exposed and which aren't, schema health against the rules we set for external developers, and quality via incidents and error rates.
• Apps Team Hub by Joe B. Everything the Apps team needs in one place — current work, incidents, bugs, ships, roadmap, and web parity.
• a11y Dashboard by Daisy. Weekly automated accessibility scans against the real product, with AI reasoning about the findings rather than just listing them. The very first run found a real bug: with a dialog open, you could tab into the page behind it.
• Community Radar by Esther. An error radar for the workers behind Buffer's Community feature, with week-over-week observations.
• Quality Dashboard by Ross. A review tool for the Advocacy team: pull conversations, set your review categories, and send feedback to a teammate. Claude can run its own analysis first if a lead wants an AI-informed starting point.
• Cleanup entities dashboard by Dinos. Inspect and requeue failed deletion workflows directly in HQ, instead of checking Datadog, checking the database, and running a script. A daily job posts to our alerts channel when something fails, with a link straight to the fix.
• Webhook subscription status by Dinos. Shows whether a channel's webhook subscription is live, with actions to check or re-subscribe. It matters because subscriptions can drop on a platform's side without us noticing.
• Mobile Beta Testers by Amanda. Aggregates beta usage by app build and reconciles it against product surface areas, sorting testers into power users, new testers, and people who haven't tested yet. It drafts the right follow-up for each segment and captures the responses.
• Jars by Abeeb. Tracks four of our tech debt jars — legacy files, type suppressions, feature flags, and TODOs — as they empty.

5. Things we couldn't see about finances

• AI spend dashboard by Jenny. Brings AI spend across six vendors and our stipend onto one consistent basis, separating committed from usage-based cost. It produced our first company-wide 2026 forecast.
• Financial consolidation by Kyle. Produces Buffer's consolidated financials for any month back to December 2012, reading both legal entities nightly from Xero and handling the currency translation and intercompany eliminations. It replaces an external tool we were paying for.
• Datadog cost dashboard by Miguel. A live dashboard and notebook drilling into where our Datadog spend goes, so nobody scrambles for detail before a contract negotiation. Visible to every teammate.
• Media Storage by Steven. How our production media bucket grew each month — files, bytes, and what they cost to keep — for the largest and least visible thing we store. Building it meant querying inventory data with Athena and shipping it to Cloudflare via Lambda, neither of which he'd done before.
• Lifecycle Experiments by Maggie. Every lifecycle experiment we're running has a live effect on the primary metric, plus a catalog of everything we've run since October 2024 with hypotheses and learnings. It also surfaces the patterns that emerge across all of them.
• Churn and revenue exploration by Steve. A dashboard connecting churn data with revenue data to understand how well our cancel-deflection flows actually work, and to pave the way for better-connected insights.
• Bid Radar by Pierre. Finds keywords Buffer should bid on in Google Ads, based on how we're already performing organically.

6. Limits customers ran into

• Text-to-image builder by Sofía. Turns a written post into an image post right in the composer — type presets, a small palette, automatic alt text, and Shift+Enter for a new slide, so text carousels work. In 2015, we built Pablo with the same pitch: that MVP took about a month; this one took a week.
• Carousels composer by José. A carousel editor for images and text, meant to live as a tab in the Create space. Built and demoed but deliberately not deployed, so it can reuse Sofía's UI elements, and the two features don't diverge.
• Mark a comment as spam by Diego. Hides a spam comment from Community for everyone in the organization, removes it from the comment score, and drops it into a reversible hidden-comments section. It also lays the groundwork for detecting repeat spammers automatically.
• WordPress plugin by Arek. A free, open-source plugin that schedules WordPress content through Buffer and keeps posts in sync with the queue. Arek works on our API internally, and building against it as an outsider produced a real proposal: a dry-run option, so you learn about a character limit before you publish, not after.
• Postwise by Iggy. Reads your own post history and tests each pattern — media type, frequency, day of week, themes — against a held-out 30%, then tells you plainly when a pattern doesn't survive. Below 100 posts a year, personalized day-of-week advice survives 32% to 46% of the time, which is worse than a coin flip.
• Buffer Pulse by Amaan. One headline score for how healthy your social habits are, built from the streaks, goals, and channel health we already track and measure over six months. Built locally for now, because storing the historical data it needs isn't worth the cost until the feature is real.
• Apple TV dashboard by Andy. A single-screen overview of your Buffer organization on tvOS, inspired by the dashboards that used to hang in the old Twitter and Pinterest offices.
• Discord streak bot by Andy. Surfaces your Buffer posting streak as a Discord role, built on a self-imposed one-hour challenge.
• Buffer Zap by Miguel. A minimalist macOS menu bar app for sharing to Buffer, built on the public API.
• BufferDrop by Mau. Drag a file onto this desktop app, and it becomes a draft post in Buffer, or point it at a folder, and anything added there does the same. Building it took him through the TUS upload protocol, Tauri, Rust, and macOS app signing, all new to him.
• Buffer Wrap by Abeeb. A quarterly summary of a customer's content and the fun highlights they might have missed.
• Post to Template by Ismail and Amaan. Paste a link to any high-performing social post and get a reusable Buffer template out of it.
• Ideas list view by Rathes. A simple list view for Ideas with the editor opening in a panel, which makes tapping through and updating them much faster than the gallery or board.

7. Friction in building Buffer itself

• Snag by Ismail. Spot a UI issue while using Buffer, annotate it in place, and Snag captures the screenshot and context, files the Linear ticket, and hands it to Roo, which opens a pull request. With the extension installed, you can also see captures made by anyone else on the team, in the product.
• Manual canaries dashboard by Peter. Diverting real traffic to an unmerged pull request used to mean posting GitHub comments in exactly the right syntax at exactly the right time. Where one codebase affects 21 deployments, engineers used to post 21 separate comments and hope — there's now a single slider.
• Mocktail by Andreas. Lets admins configure pretend accounts and channel setups and save them as profiles, so you can test how a feature behaves with no channels or a great many without creating an account for each. Kept on a branch environment rather than merged, because he didn't want to risk app load times for customers before he was certain.
• Sound Buffet by Andreas. A browsable library of quirky, self-recorded UI sound effects, with previews, downloads, and a ready-to-paste snippet.
• Help Scout connector by Andreas. Improvements to the shared connector that several other Build Week projects then used.
• StarDeps Valley by Bruno. Dependency maintenance as a Stardew Valley-style garden: our two most critical repositories are rendered as garden beds, and claiming a plant triggers a workflow that opens the pull request for you. Boring, necessary work, made into something you might want to do.
• Roo HQ by Kateryna. A home for our agents platform, showing what Roo is doing now, full session history across Slack, GitHub, Linear, and scheduled routines, and detailed views with reasoning, tool use, costs, and traces.
• Ambient mode for Roo by Kateryna. Roo can now be given standing instructions and decide to engage on its own without being tagged. It also reads a burst of thread replies together and answers once, instead of treating each as a separate task.
• Coach-a-Roo by Adam. Anyone can review how Roo, our internal agent, would have handled a real past support conversation, rate it, and leave feedback. That crowdsourced feedback then improves the persona.
• Clockwork by Kateryna. One place for Buffer's scheduled automations — the reports, reminders, and digests that run on their own.
• Buffer Skills proposal by Kateryna. Research into how we should organize the Claude skills currently scattered across repos and personal setups. Deliberately written up as a proposal rather than shipped, because it needed alignment, and she didn't want to interrupt anyone's Build Week.
• Popcorn motion system by Héctor. One shared set of timings and easings, with dialogs, drawers, toasts, loaders, and page headers all moved onto it across 30 pull requests. Loading indicators now wait 300 milliseconds before appearing and stay for at least 400, so nothing flashes on a quick load. (Popcorn is what we call our design system.)
• Popcorn Kernel by Andy. A condensed version of our design system built for product previews on the marketing site, so we can show real animating components instead of static screenshots. It adjusts typography, spacing, and radii for small-scale views rather than just scaling everything down.
• Popcorn RichTextField by Mw. A first version of a modular, reusable rich text component supporting a large set of Markdown features plus mentions and character limits, with plugins consumers can include, edit, or overwrite.
• Popcorn styler skill by Sofía. A Claude skill that applies our product's look and feel, so anyone can mock up something that looks like Buffer.
• Front-end standards for agents by Mw. A composition rule file that loads whenever an agent opens a React file, plus a new lint rule, tested with two different models to confirm that both read it. The decision behind it: invest in linting and context files, because evals are expensive and often don't pay off.
• Feedback widget improvements by Brandon. Our in-product feedback form now asks more pointed questions and has a lightweight spam check, which should improve the quality of what we get.
• Feedback routing and moderation by Brandon. Routes in-product feedback to our central hub instead of a clunky Google Sheet we can now decommission, plus a recurring routine that reviews new suggestions, merges obvious duplicates, removes spam, and hands him a one-click list of the rest.
• Inbox & Chill by Brandon. A macOS menu bar app that puts Slack, Linear, Notion, GitHub, Sentry, Apple Mail, and your terminal sessions into one queue you can empty. Built while waiting on feedback for his main project.

8. Builds for personal workflows

• The Evoker Assistant by Jess. A second brain pairing Claude with Obsidian to replace Notion and Dropbox Paper and keep her notes local. She realized four half-days weren't enough to migrate everything, so she deliberately built the framework and left the import for later.
• A work vault that actually works by Mel. She set out to build a task tracker, built it, and while building it, concluded it wouldn't solve her problem — most of what she needs to track is too small to be worth opening an app for. She came out with a much better vault and a clear read on what the tool is and isn't for.
• Recipe capture skills by Mel. Two skills that turn screenshots and links into searchable recipe notes, each with a full ingredient list and a line on how to adapt it so her kids will actually eat it. A day-four pivot that finished the week with something she'll use.
• Mark, an exam technique coach by Edu. A tutoring agent for A Level Psychology that picks a real past-paper question, reads a photo of a handwritten answer, marks it against the marking scheme, and suggests improvements without writing the answer. He built a bank of 70 questions and used evals for the first time to check the marking against real examiner notes.
• Trip Photos Autopilot by Esther. Scans a Drive folder of travel photos, detects duplicates, converts files within Instagram's format rules, writes captions, and sends each post to Buffer 24 hours before it publishes so there's time to review.
• Create from Everyday Actions by Cheryl. A skill that reads her own Slack messages, Linear issues, and troubleshooting investigations, finds the human dynamic under a work problem, and saves a parenting-themed draft to her Threads queue. When she asked Claude to evaluate it, it noticed she'd edited every draft, found the pattern, and suggested saving that as a learning.
• Time-off return routine by Cheryl. Scans our time-off channel and DMs people returning from leave in the next three days to ask whether they'd like to share how it went, with three ways to answer.
• Blog-to-social routine by Rathes. A Claude routine that checks his blog weekly and drafts the social posts for four channels. It turned out far simpler than he expected, which freed the rest of his week.
• A revamped second brain by Mw. Cleaned, sorted, and rebuilt around atomic notes, with AI allowed to find missing links and tags but never to edit note contents.
• Downstream by Juliet. Measures rather than estimates her own AI usage, and has already turned up something she didn't expect: 45% of hers came from subagents she'd spawned.

9. Just for fun!

• Bufferverse by Tami. Buffer as a universe: all 78 of us are stars, and the sky lights up by nine things we might share, from team and tenure to Chinese zodiac and birth order. Click any star to see what you have in common with that person, and there's a daily "guess the star" quiz with the same puzzle for everyone.
• Houseswap by Pierre. An internal home exchange — list your place, manage availability, browse where other Bufferoos live, and send or accept swap requests, with airports, weather, and budget pulled in. During the week, it grew an entire pet-sitting feature and local guides, both from feedback.
• Confetti by Steve. Buffer crossed $26 million in ARR on 18 August, in the middle of Build Week. Replay the moment, and click the emoji as often as you like.
• Creative Brain Games by Kate. Seven small games for flexing creative thinking, all based on real, researched exercises and deliberately unrelated to work: sketch as many things as you can in a minute, find as many uses as you can for one ordinary object, chain two unrelated words together. Her reason: it matters that we keep practicing creative thinking as we pass more and more over to AI.
• TWST Leaderboard by Juliet. Counts everyone's contribution live during a Team Wide Support Time session and throws confetti at the end. It started as a throwaway artifact for Build Week itself and became a tool we'll use at retreats.
• Team of Creators board by Diego C. Connect your Buffer account, and you get a cell showing what you published this week, plus a level, which is simply how many posts you published. Posts published natively count too, not only the ones sent through Buffer.
• Roll by Hailley. Roll two dice, get a true fact about Buffer. Anyone can add their own.
• Guess the Pets of Buffer by Hailley. Guess which teammate each Buffer pet belongs to.
• Buffer Jeopardy by Hailley. Host a Jeopardy-style trivia game for an in-person team activity, built from a CSV of clues.
• Screensaver by Hailley. The Buffer logo, bouncing. Wait for it to hit a corner.
• Hop-a-Roo by Kateryna. A one-key hopping game with our agent as the character, and a leaderboard for the whole team.
• Full House by Suzanne. Compares Buffer's customer count to stadiums, cities, and Olympic Park villages, because someone had to.
• Month Wrap by Daisy. Your last 30 days in Buffer, as a short animated wrap in seven slides.
• Drop by Kateryna. Run a swag drop: pick the merch, open it to the team, and let people claim theirs.
• Value-led growth page by Mike E. A marketing page that externalizes how we think about generosity as a strategy, so other businesses can take inspiration from it. Previewed, not live yet.

We're incredibly impressed with everything our team built, and we hope some of it makes its way to you. Reach out any time on social media or in our Community with any thoughts or questions!

More resources on how we work

• Our Team Built 17 Improvements to Buffer This Week, Here's The Recap
• We Added a $250 AI Tools Stipend for the Buffer Team
• The AI Mindset Shift: How I Use AI Every Day
• 35 Side Projects From the Buffer Team
• Build Week at Buffer