Replicas growth playbook

Write the answer in the FAQ and as a short table above it: three task types (a short automated job, a typical ticket, a long multi-repo task), their minutes from customer data, the compute cost on Small and Large, and how many of each the $50 Developer plan and the $100 free usage cover (the $100 buys 12,500 Large minutes, about 1,000 Parrot-length jobs or 85 Knowunity-length tasks). State clearly that model costs are separate and billed by the customer's own plan. Then add a three-input calculator (tasks a week, minutes a task, workspace size) that recommends Developer or a Team step, and copy the same answer into /llm-info and the vs pages so assistants and coding agents quote it.

Usage pricing stalls deals when buyers cannot translate it into a budget; a per-task figure in cents removes the biggest objection to trying cloud agents, and it contrasts with rivals whose quota and credit units are harder to compare. It also gives the Founding Growth hire an answer AI assistants can cite when asked what cloud coding agents cost.

Most tasks cost cents of compute. Replicas bills per workspace minute while the workspace is awake; model usage stays on your own plan or key.

From customer usage in July 2026:
- Short automated job (about 11 minutes, like Parrot's review jobs): about $0.05 on Small, $0.09 on Large
- Typical ticket (about 100 minutes, like Moda's tasks): about $0.40 on Small, $0.85 on Large
- Long multi-repo task (about 140 minutes, like Knowunity's on Large): about $1.15

The $50 Developer plan covers 12,500 Small or 6,250 Large minutes. Your $100 of free usage covers about 1,000 short jobs or 85 long tasks on Large. Sleeping and archived workspaces cost nothing.

• Buyers anchor on the lowest figure and feel misled by idle awake time

  Show median and 90th percentile, and explain that sleeping workspaces cost nothing.

• Model costs dwarf compute and the table looks incomplete

  Say plainly that model usage stays on the customer's own plan, and link the existing model pricing explainers.

• Your websitePricing page (checked November 10, 2026): Developer "$50/mo" with $50 of usage included, Team "$200+/mo" in $100 steps, Enterprise custom; per-minute rates Small (2 vCPU, 8 GB) $0.004 included and $0.006 overage, Large (4 vCPU, 16 GB) $0.008 and $0.012, Ultra (8 vCPU, 32 GB) $0.016 and $0.024; "Model costs are not included". Eight FAQ questions, but only the first has an answer in the page: "What does a typical task cost?", "Is there a free trial?", "What counts as usage?", "What happens if we use more than our plan includes?", the per-seat question and the Devin and Cursor comparisons have no answer text (checked three times). No calculator; "$100 of free usage, no card required" appears only in the closing call to action.Source
• Your website"For AI assistants" page (/llm-info): a pricing summary but no typical cost or minutes per task; customer usage for July: Parrot "23.5k automated jobs" and "270k+ minutes", Moda "4.7k jobs" and "500k+ minutes".Source
• Your websiteKnowunity story: an AI learning platform for "more than 30 million students"; 30+ engineers on Replicas in July; a target of 25% of PRs and MRs from agents in month one, now 50%; "2×" engineering velocity; bug-fixing time down two-thirds; "2.8k" cloud agents and "400,000+ minutes" in July on 4 vCPU, 16 GB workspaces; daily analytics of usage, compute, token spend and PR output by source and engineer.Source
• Your websiteReplicas vs Devin (updated October 25, 2026): Jira works "only as an agent tool, not a trigger" while Devin starts work from Jira and Teams; Devin has SOC 2 Type II "since September 2024"; the page says Devin may fit better for Jira or Azure DevOps triggers or "SOC 2 Type II today"; it gives no example cost per task for Replicas.Source
• HiringFounding Growth (Y Combinator jobs, San Francisco, $150K-$250K, 0.5-2% equity): owns the funnel from "signup to onboarding to power user"; SEO, GEO, comparison pages and docs so AI assistants and coding agents recommend Replicas; X is "the biggest growth channel" with constant launches; in-person events for founders in SF; the company reports "six figures in ARR", tens of thousands of PRs and millions of usage minutes a month, and aims for eight figures; team of 2.Source

Write a two- to four-sentence answer for every question, using facts already on the site (no per-seat fee, per-minute billing while awake, overage rates, Devin's per-member and per-seat plans, model costs on the customer's own plan). Render answers in the HTML, add FAQ structured data, and add "Start with $100 of free usage, no card" under the Developer and Team prices. Fill the Enterprise FAQ the same way.

Unanswered questions on a pricing page read as hidden costs, and the free usage is the strongest reason to try today. Answers in the HTML are also what search engines and AI assistants can quote.

• Competitor pricing in answers goes stale

  Date the comparison answers and link the dated vs pages.

• Your websitePricing page (checked November 10, 2026): Developer "$50/mo" with $50 of usage included, Team "$200+/mo" in $100 steps, Enterprise custom; per-minute rates Small (2 vCPU, 8 GB) $0.004 included and $0.006 overage, Large (4 vCPU, 16 GB) $0.008 and $0.012, Ultra (8 vCPU, 32 GB) $0.016 and $0.024; "Model costs are not included". Eight FAQ questions, but only the first has an answer in the page: "What does a typical task cost?", "Is there a free trial?", "What counts as usage?", "What happens if we use more than our plan includes?", the per-seat question and the Devin and Cursor comparisons have no answer text (checked three times). No calculator; "$100 of free usage, no card required" appears only in the closing call to action.Source
• Your websiteEnterprise page: "8M+ minutes of agent runtime" and "290k+ agents run"; Replicas cloud, dedicated single-tenant or self-hosted deployment; SCIM, audit log, DPA and HIPAA BAA; SOC 2 "Type I" with "Type II in progress"; the pricing FAQ has no visible answer.Source
• Your websiteReplicas vs Devin (updated October 25, 2026): Jira works "only as an agent tool, not a trigger" while Devin starts work from Jira and Teams; Devin has SOC 2 Type II "since September 2024"; the page says Devin may fit better for Jira or Azure DevOps triggers or "SOC 2 Type II today"; it gives no example cost per task for Replicas.Source

Run a time-boxed concierge: a homepage and X banner for Conductor Cloud users, 20 founder-led 30-minute migration slots, migration done for them with the existing prompt and CLI, and extra usage credits for teams who move before the shutdown. Post daily on X with a countdown and one migrated team's before-and-after, and follow up after November 20 with teams who exported data but have not moved.

Displaced users of a closing product are the highest-intent audience in the category, and the concepts map almost one to one (Cloud Computer to environment, routines to automations).

• Appearing to celebrate a rival's shutdown

  Keep the tone practical: export steps first, Replicas second.

• Users expect chat history import

  Say upfront that chat history cannot be imported, as the guide does.

• PressInfoWorld: Meta acquired Conductor; its cloud service becomes "read-only" on November 13 and shuts down on November 20, 2026, when deletion of cloud user data begins; the local app gets new models and critical fixes until early 2027, then winds down.Source
• Your websiteBlog "Migrate from Conductor Cloud" (November 4, 2026): maps Conductor concepts to Replicas (Cloud Computer to environment, install script to warm hook, routines to automations), offers a migration prompt that recreates the setup with the Replicas CLI, and help by email to the founders; Replicas "can't import chat history".Source
• Your websiteSitemap (120 URLs, checked November 10, 2026): 61 resource pages, including 16 "alternatives" pages (among them Conductor, Devin, Factory and Codex), 8 "vs" pages and pricing explainers for three rival tools; 28 blog posts, including guides on using PostHog, ClickHouse, Mercury, Vanta and Context Dev with Replicas; 5 customer stories; 5 integration pages; /yc and /trading pages; no changelog, templates, referral or partner page.Source

Define each metric once (agent minutes, workspaces, agents, PRs opened), pull them from the analytics into one live counter component, and use it on the homepage, Enterprise page, /llm-info and the YC page with an "as of" date. Post the monthly milestone on X.

Consistent, dated numbers read as trustworthy, and a live counter turns growth into a recurring launch moment for the X-led channel.

• Showing the counter invites scrutiny if growth slows

  Show cumulative totals and pick metrics that reflect real work, such as PRs opened.

• Your websiteSeed announcement (October 13, 2026): $4M from Y Combinator, Flex Capital, Founders' Co-op, Pioneer Fund and others; usage passed 1 million agent minutes in May, 2 million in July and 5 million in September, and was "nearing 7 million" three weeks later; agents have run in "over 1 million workspaces"; Mintlify and Knowunity named as customers.Source
• Your websiteEnterprise page: "8M+ minutes of agent runtime" and "290k+ agents run"; Replicas cloud, dedicated single-tenant or self-hosted deployment; SCIM, audit log, DPA and HIPAA BAA; SOC 2 "Type I" with "Type II in progress"; the pricing FAQ has no visible answer.Source
• HiringFounding Growth (Y Combinator jobs, San Francisco, $150K-$250K, 0.5-2% equity): owns the funnel from "signup to onboarding to power user"; SEO, GEO, comparison pages and docs so AI assistants and coding agents recommend Replicas; X is "the biggest growth channel" with constant launches; in-person events for founders in SF; the company reports "six figures in ARR", tens of thousands of PRs and millions of usage minutes a month, and aims for eight figures; team of 2.Source

Generate share images automatically for every page type: homepage, blog posts (title plus date), customer stories (headline metric and logo), vs and alternatives pages, and the pricing page (the per-task table). Validate with each network's card preview and keep a fallback image for any page without one.

Link cards with images take far more space and attention in feeds and Slack unfurls, which is where engineering teams share tools with each other.

• Generic images get ignored

  Put the page's key number in the image.

• Researchprogrammable.info research (November 10, 2026): real visitors pass Core Web Vitals (LCP p75 1.7 s, 19% faster than before; INP 79 ms; CLS 0.01); the homepage has no social share image; the Show HN post (November 2025) got 1 point and no comments; the iOS app has 1 rating; the GitHub organisation has 2 public repositories and 3 followers; no public changelog or roadmap was found.Source
• HiringFounding Growth (Y Combinator jobs, San Francisco, $150K-$250K, 0.5-2% equity): owns the funnel from "signup to onboarding to power user"; SEO, GEO, comparison pages and docs so AI assistants and coding agents recommend Replicas; X is "the biggest growth channel" with constant launches; in-person events for founders in SF; the company reports "six figures in ARR", tens of thousands of PRs and millions of usage minutes a month, and aims for eight figures; team of 2.Source

After repos connect, suggest three starter tasks generated from the repository itself (a failing test, a stale dependency, a missing README section or open issues labelled good first issue), run one by default, and show the PR with its video proof. Track the funnel: signup, agent connected, repo connected, first workspace, first PR, first merge, second week of use. Email or Slack a nudge at each drop-off.

Usage-based products grow when users reach a real result fast; a merged PR in the first session proves the whole loop and starts the habit.

• Auto-run tasks open unwanted PRs

  Ask before opening the PR and label it as a starter task.

• Your websiteDocs quickstart ("about 5 minutes"): connect a coding agent subscription or API key with the CLI, connect GitHub or GitLab, add repositories (an environment is created for each), optionally connect Slack or Linear, then "describe a small task" to start the first workspace.Source
• HiringFounding Growth (Y Combinator jobs, San Francisco, $150K-$250K, 0.5-2% equity): owns the funnel from "signup to onboarding to power user"; SEO, GEO, comparison pages and docs so AI assistants and coding agents recommend Replicas; X is "the biggest growth channel" with constant launches; in-person events for founders in SF; the company reports "six figures in ARR", tens of thousands of PRs and millions of usage minutes a month, and aims for eight figures; team of 2.Source
• Your websiteKnowunity story: an AI learning platform for "more than 30 million students"; 30+ engineers on Replicas in July; a target of 25% of PRs and MRs from agents in month one, now 50%; "2×" engineering velocity; bug-fixing time down two-thirds; "2.8k" cloud agents and "400,000+ minutes" in July on 4 vCPU, 16 GB workspaces; daily analytics of usage, compute, token spend and PR output by source and engineer.Source

When a workspace org reaches 80% of included usage, show in-app and by email: minutes used, projected month-end, cost with overage versus the next $100 step, and a one-click upgrade. For Developer accounts with several members, suggest Team for the shared Slack channel and static egress IP. Report monthly expansion revenue from these prompts.

Usage-based revenue grows with customers' agent adoption; showing the cheaper option builds trust and moves teams up before a surprise bill.

• Lower short-term overage revenue

  Committed steps are steadier revenue and reduce churn from bill shock.

• #1 Answer "What does a typical task cost?" on the pricing page with customers' own published numbers

• Your websitePricing page (checked November 10, 2026): Developer "$50/mo" with $50 of usage included, Team "$200+/mo" in $100 steps, Enterprise custom; per-minute rates Small (2 vCPU, 8 GB) $0.004 included and $0.006 overage, Large (4 vCPU, 16 GB) $0.008 and $0.012, Ultra (8 vCPU, 32 GB) $0.016 and $0.024; "Model costs are not included". Eight FAQ questions, but only the first has an answer in the page: "What does a typical task cost?", "Is there a free trial?", "What counts as usage?", "What happens if we use more than our plan includes?", the per-seat question and the Devin and Cursor comparisons have no answer text (checked three times). No calculator; "$100 of free usage, no card required" appears only in the closing call to action.Source
• Your websiteSupport page: email to the founders' address with screenshots; no stated response time, community or in-app chat; a shared Slack support channel comes only with Team and Enterprise plans.Source

Build Jira as a trigger (assign an issue to Replicas, get a PR and a status change, as with Linear), then Bitbucket repos. Add a "request an integration" form on the integrations page to rank what comes next (PagerDuty, Teams). Update the Devin and alternatives pages when each ships.

Larger engineering teams often run on Jira; a missing trigger removes Replicas from their shortlists before price is even compared.

• Engineering time pulled from core platform

  Ship Jira first and build Bitbucket only if requests justify it.

• Your websiteReplicas vs Devin (updated October 25, 2026): Jira works "only as an agent tool, not a trigger" while Devin starts work from Jira and Teams; Devin has SOC 2 Type II "since September 2024"; the page says Devin may fit better for Jira or Azure DevOps triggers or "SOC 2 Type II today"; it gives no example cost per task for Replicas.Source
• Your websiteIntegrations page: Slack, Linear, GitHub, GitLab and Sentry, plus the app, API and CLI; no Jira, Bitbucket, PagerDuty or Microsoft Teams, and no way to request an integration.Source
• Competitor researchTembo homepage: "Cloud platform for running AI agents in complete, shareable software environments"; triggers from GitHub, GitLab, Bitbucket, Linear, Jira, Sentry, PagerDuty, Slack and webhooks; public preview links; SOC 2 Type II and self-hosted or VPC deployment.See the research Source

Point the existing changelog automation at a public /changelog page with an RSS feed: one entry a week, written by an agent and checked by a founder, each line linking to docs. Label it "written by a Replicas automation" and link the software-factory post. Post the weekly highlights on X.

A visible weekly cadence reassures buyers worried about a young vendor, and an agent-written changelog demonstrates the product doing real work.

• Agent-written entries leak internal details

  Founder review before each entry publishes.

• Your websiteBlog "How We Run Our Software Factory" (October 22, 2026): a #triage Slack channel where an agent turns bugs into Linear tickets; crons for Postgres query tuning, Linear cleanup, a "weekly changelog", stale PR cleanup and a code simplification audit; an agent runs the CRM; the team reviews video or screenshot proof instead of diffs; an autofix agent makes CI green before the PR opens.Source
• Your websiteSitemap (120 URLs, checked November 10, 2026): 61 resource pages, including 16 "alternatives" pages (among them Conductor, Devin, Factory and Codex), 8 "vs" pages and pricing explainers for three rival tools; 28 blog posts, including guides on using PostHog, ClickHouse, Mercury, Vanta and Context Dev with Replicas; 5 customer stories; 5 integration pages; /yc and /trading pages; no changelog, templates, referral or partner page.Source
• Researchprogrammable.info research (November 10, 2026): real visitors pass Core Web Vitals (LCP p75 1.7 s, 19% faster than before; INP 79 ms; CLS 0.01); the homepage has no social share image; the Show HN post (November 2025) got 1 point and no comments; the iOS app has 1 rating; the GitHub organisation has 2 public repositories and 3 followers; no public changelog or roadmap was found.Source

Ship a gallery of ten automation templates, each with the prompt, trigger, schedule, workspace size, expected minutes and cost per run: daily query tuning, stale PR cleanup, failing CI fix, Sentry issue to PR, weekly changelog, dependency bumps, PR review, flaky test hunt, docs drift check, Linear triage. One click installs it into the user's environment. Each template gets a public page that ranks for its task.

Automations create recurring, predictable usage without anyone typing a prompt, which is how a usage-based product grows per account; public template pages also capture searches for each task.

• Automations open noisy PRs and teams switch them off

  Default to weekly schedules and a cap on open PRs per template.

• #6 Make the first session end in a merged PR, and measure signup to first merge

• Your websiteBlog "How We Run Our Software Factory" (October 22, 2026): a #triage Slack channel where an agent turns bugs into Linear tickets; crons for Postgres query tuning, Linear cleanup, a "weekly changelog", stale PR cleanup and a code simplification audit; an agent runs the CRM; the team reviews video or screenshot proof instead of diffs; an autofix agent makes CI green before the PR opens.Source
• Your website"For AI assistants" page (/llm-info): a pricing summary but no typical cost or minutes per task; customer usage for July: Parrot "23.5k automated jobs" and "270k+ minutes", Moda "4.7k jobs" and "500k+ minutes".Source
• Your websiteSitemap (120 URLs, checked November 10, 2026): 61 resource pages, including 16 "alternatives" pages (among them Conductor, Devin, Factory and Codex), 8 "vs" pages and pricing explainers for three rival tools; 28 blog posts, including guides on using PostHog, ClickHouse, Mercury, Vanta and Context Dev with Replicas; 5 customer stories; 5 integration pages; /yc and /trading pages; no changelog, templates, referral or partner page.Source

Once a month, invite 20-30 founders and engineering leads (YC batch first). Each brings three backlog tickets; Replicas runs them in parallel on the big screen while a short talk covers the software factory. Everyone leaves with PRs, an account with credits and a recording to share on X.

Seeing your own tickets solved is far more persuasive than a demo repo, and the shared recordings feed the X channel.

• Private code on a public screen

  Show only the team's own screen to them; use a demo repo for the big screen.

• HiringFounding Growth (Y Combinator jobs, San Francisco, $150K-$250K, 0.5-2% equity): owns the funnel from "signup to onboarding to power user"; SEO, GEO, comparison pages and docs so AI assistants and coding agents recommend Replicas; X is "the biggest growth channel" with constant launches; in-person events for founders in SF; the company reports "six figures in ARR", tens of thousands of PRs and millions of usage minutes a month, and aims for eight figures; team of 2.Source
• Your website/yc page "The YC Deal": current-batch companies get monthly usage credits for three months, dedicated onboarding and migration, and direct founder access; alumni get one month of credits and onboarding; claimed through a YC verification link.Source

Give both sides usage credits when a referred team becomes a paying org. Put a referral link in the app's billing page and the monthly usage email, and a "refer a batchmate" line on the YC page. Ask angels and investors for a short intro list to their portfolio CTOs.

Startup engineering leaders trust peers more than ads, and credits cost Replicas only compute at its own included rate.

• Credit farming

  Pay the credit only after the referred org's first paid month.

• Your website/yc page "The YC Deal": current-batch companies get monthly usage credits for three months, dedicated onboarding and migration, and direct founder access; alumni get one month of credits and onboarding; claimed through a YC verification link.Source
• Your websiteSeed announcement (October 13, 2026): $4M from Y Combinator, Flex Capital, Founders' Co-op, Pioneer Fund and others; usage passed 1 million agent minutes in May, 2 million in July and 5 million in September, and was "nearing 7 million" three weeks later; agents have run in "over 1 million workspaces"; Mintlify and Knowunity named as customers.Source
• Your websiteSitemap (120 URLs, checked November 10, 2026): 61 resource pages, including 16 "alternatives" pages (among them Conductor, Devin, Factory and Codex), 8 "vs" pages and pricing explainers for three rival tools; 28 blog posts, including guides on using PostHog, ClickHouse, Mercury, Vanta and Context Dev with Replicas; 5 customer stories; 5 integration pages; /yc and /trading pages; no changelog, templates, referral or partner page.Source

Rewrite the Parrot, Moda and Utility Profit headlines around their numbers, add one quote and the key metric to each card on the index, and tag each story by trigger (Slack, Linear, automations) and team size. Ask three logo-wall customers for a short story built from their own analytics: PR share, minutes, cost per PR.

Engineering leaders compare outcomes; a number in the headline is what gets read and quoted, by people and by AI assistants.

• Customers decline to share numbers

  Offer ranges or percentages instead of absolute figures.

• Your websiteCustomers page: five stories (Knowunity, Parrot, Moda, Utility Profit "ships code across all departments", Vela); only two headlines carry a number (Knowunity "doubled engineering velocity", Vela "3×" faster reviews); a logo wall of 16 companies including Mintlify, Composio and Tensorlake; no quotes on the index.Source
• Your websiteHomepage: "Give Your Backlog to Replicas"; runs six agent harnesses, "each in its own Linux VM", started from Linear, Slack, GitHub, GitLab or the dashboard; Knowunity "50% of pull requests produced by cloud agents", Parrot "23.5k automated cloud-agent jobs in July", Moda "100% of its engineering team"; "$100 of free usage, no card required"; "Book a demo" opens a 30-minute calendar booking; footer links to alternatives pages; "SOC 2 Type I (Type II in progress)".Source

Publish the Type II audit window and expected report date, answer the SOC 2 FAQ, and build a downloadable security pack: architecture (one VM per task, destroyed after), data retention options, bring-your-own model credentials, secret sync, network controls, and a pre-filled standard security questionnaire. Offer bridge letters and a security call with the CTO for enterprise deals until Type II arrives.

Security review is where young vendors lose enterprise deals; a ready pack and a dated Type II plan turn "not yet" into "on this date".

• Report date slips

  Publish a window rather than a day and update it.

• Your websiteEnterprise page: "8M+ minutes of agent runtime" and "290k+ agents run"; Replicas cloud, dedicated single-tenant or self-hosted deployment; SCIM, audit log, DPA and HIPAA BAA; SOC 2 "Type I" with "Type II in progress"; the pricing FAQ has no visible answer.Source
• Your websiteReplicas vs Devin (updated October 25, 2026): Jira works "only as an agent tool, not a trigger" while Devin starts work from Jira and Teams; Devin has SOC 2 Type II "since September 2024"; the page says Devin may fit better for Jira or Azure DevOps triggers or "SOC 2 Type II today"; it gives no example cost per task for Replicas.Source
• Competitor researchDevin homepage: "your team's autonomous software engineer"; works with Slack, Teams, GitHub, Jira and Linear; claims SOC 2, ISO 27001 and FedRAMP; logos of large enterprises.See the research Source
• Competitor researchTembo homepage: "Cloud platform for running AI agents in complete, shareable software environments"; triggers from GitHub, GitLab, Bitbucket, Linear, Jira, Sentry, PagerDuty, Slack and webhooks; public preview links; SOC 2 Type II and self-hosted or VPC deployment.See the research Source

Make one figure the source of truth, generated by the agents and dated. Reframe the page headline around the engineering result: days of uptime without human help, workspaces used, self-repairs, compute cost per day. Post a weekly X thread from the agent journal, and end with a write-up on what broke and how agents fixed it.

The audience is engineers buying reliability, not investors; uptime and self-repair are the claims Replicas sells, and a consistent number keeps the experiment credible.

• A trading theme attracts the wrong audience

  Lead with engineering metrics and keep the simulated-money disclaimer visible.

• Your website/trading page: "An AI hedge fund run entirely by cloud agents", a 60-day paper-trading experiment with $100,000 of simulated money split between four AI models with no human help; on day 33 the fund was down and behind the S&P 500, and the page shows three different returns (-4.672% in the headline, -5.25% in the day 33 note, -5.60% at day 30).Source
• HiringFounding Growth (Y Combinator jobs, San Francisco, $150K-$250K, 0.5-2% equity): owns the funnel from "signup to onboarding to power user"; SEO, GEO, comparison pages and docs so AI assistants and coding agents recommend Replicas; X is "the biggest growth channel" with constant launches; in-person events for founders in SF; the company reports "six figures in ARR", tens of thousands of PRs and millions of usage minutes a month, and aims for eight figures; team of 2.Source

Publish an open-source skill and MCP server that lets a local agent say "run this in Replicas": it packages the current branch and task, starts a workspace via the CLI and returns the PR link. List it in the main skill and MCP directories, document it in the docs, and add a CLI login flow that grants the $100 free usage.

It meets developers inside the tool they already use at the exact moment the laptop becomes the bottleneck, and the open-source repo is a distribution channel in its own right.

• Agents send secrets with the hand-off

  Reuse the migration prompt's rule of keeping secret values out of the conversation.

• HiringFounding Growth (Y Combinator jobs, San Francisco, $150K-$250K, 0.5-2% equity): owns the funnel from "signup to onboarding to power user"; SEO, GEO, comparison pages and docs so AI assistants and coding agents recommend Replicas; X is "the biggest growth channel" with constant launches; in-person events for founders in SF; the company reports "six figures in ARR", tens of thousands of PRs and millions of usage minutes a month, and aims for eight figures; team of 2.Source
• Your websiteBlog "Migrate from Conductor Cloud" (November 4, 2026): maps Conductor concepts to Replicas (Cloud Computer to environment, install script to warm hook, routines to automations), offers a migration prompt that recreates the setup with the Replicas CLI, and help by email to the founders; Replicas "can't import chat history".Source
• ResearchGitHub organisation replicas-group ("Background coding agents for engineering teams"): two public repositories, "mographc" (6 stars) and "skill" (1 star), and 3 followers.Source
• Your websiteSeed announcement (October 13, 2026): $4M from Y Combinator, Flex Capital, Founders' Co-op, Pioneer Fund and others; usage passed 1 million agent minutes in May, 2 million in July and 5 million in September, and was "nearing 7 million" three weeks later; agents have run in "over 1 million workspaces"; Mintlify and Knowunity named as customers.Source

Email the org admin on the first of each month: PRs opened and merged by agents, share of all PRs, minutes and compute cost, cost per merged PR, top triggers and skills, and one suggestion (an automation template or a workspace size change). Make it forwardable with a share image for internal reporting.

Usage products retain when the buyer sees value regularly; cost per merged PR is the number a CTO repeats to their CEO and board.

• A bad month prompts cancellation

  Pair low months with a concrete fix and an offer of help.

• Your websiteKnowunity story: an AI learning platform for "more than 30 million students"; 30+ engineers on Replicas in July; a target of 25% of PRs and MRs from agents in month one, now 50%; "2×" engineering velocity; bug-fixing time down two-thirds; "2.8k" cloud agents and "400,000+ minutes" in July on 4 vCPU, 16 GB workspaces; daily analytics of usage, compute, token spend and PR output by source and engineer.Source
• Your websiteHomepage: "Give Your Backlog to Replicas"; runs six agent harnesses, "each in its own Linux VM", started from Linear, Slack, GitHub, GitLab or the dashboard; Knowunity "50% of pull requests produced by cloud agents", Parrot "23.5k automated cloud-agent jobs in July", Moda "100% of its engineering team"; "$100 of free usage, no card required"; "Book a demo" opens a 30-minute calendar booking; footer links to alternatives pages; "SOC 2 Type I (Type II in progress)".Source

Add an in-app help button that opens a thread handled first by a Replicas agent trained on the docs (answers setup questions, reproduces bugs in a workspace, files a ticket), with a founder taking over when needed. Publish a response target (for example one business day for Developer, four hours for Team) on the support and pricing pages.

Fast, visible support lowers the risk of choosing a two-person vendor, and agent-first support is another live demo of the product.

• Agent gives wrong answers

  Label agent replies and let users ask for a human.

• Your websiteSupport page: email to the founders' address with screenshots; no stated response time, community or in-app chat; a shared Slack support channel comes only with Team and Enterprise plans.Source
• Your websiteBlog "How We Run Our Software Factory" (October 22, 2026): a #triage Slack channel where an agent turns bugs into Linear tickets; crons for Postgres query tuning, Linear cleanup, a "weekly changelog", stale PR cleanup and a code simplification audit; an agent runs the CRM; the team reviews video or screenshot proof instead of diffs; an autofix agent makes CI green before the PR opens.Source

Ship guided review on iOS (summary, most important file first, ask the agent for a fix, approve and merge), with push notifications when an agent's PR is ready. After a user's third mobile merge, show the native rating prompt. Promote "merge from your phone" clips on X.

Reviewing from anywhere makes agents part of the day, not a desk task; ratings and reviews help the app get found in the App Store.

• Merging from a phone without enough care

  Respect branch protections and show the agent's test and video proof first.

• Your websiteBlog "Introducing Replicas Review" (October 21, 2026): one place to read, discuss, fix and merge GitHub PRs and GitLab MRs; guided reviews; an agent pushes fixes from a chat beside the diff; an inbox, stacked PRs and keyboard shortcuts; available on web and desktop.Source
• Researchprogrammable.info research (November 10, 2026): real visitors pass Core Web Vitals (LCP p75 1.7 s, 19% faster than before; INP 79 ms; CLS 0.01); the homepage has no social share image; the Show HN post (November 2025) got 1 point and no comments; the iOS app has 1 rating; the GitHub organisation has 2 public repositories and 3 followers; no public changelog or roadmap was found.Source

Send each guide to the partner's developer relations team with an offer: a co-written post, a listing in their integrations or partner directory, a joint X post and credits for their users. Turn the guides into an /integrations section for "tools your agents can use". Start with the two partners whose users overlap most with Replicas' customers.

Partner directories and co-posts bring trusted, targeted traffic, and the guides are already written.

• Partners are slow to respond

  Ask investors and angels with ties to those companies for intros.

• Your websiteSitemap (120 URLs, checked November 10, 2026): 61 resource pages, including 16 "alternatives" pages (among them Conductor, Devin, Factory and Codex), 8 "vs" pages and pricing explainers for three rival tools; 28 blog posts, including guides on using PostHog, ClickHouse, Mercury, Vanta and Context Dev with Replicas; 5 customer stories; 5 integration pages; /yc and /trading pages; no changelog, templates, referral or partner page.Source
• Your websiteIntegrations page: Slack, Linear, GitHub, GitLab and Sentry, plus the app, API and CLI; no Jira, Bitbucket, PagerDuty or Microsoft Teams, and no way to request an integration.Source

Each quarter, publish aggregated, anonymised data: merge rate and median minutes by task type, trigger and harness; compute cost per merged PR; how automations compare with hand-assigned tasks; trends quarter on quarter. Release a public dashboard and a PDF, brief developer press, and invite customers to benchmark themselves against the medians in-app.

Original data earns links, press and AI-assistant citations that comparison pages cannot, and it positions Replicas as the neutral layer above all harnesses.

• Harness vendors dispute the findings

  Publish the method, sample sizes and caveats, and avoid ranking claims the data cannot support.

• Customer data sensitivity

  Aggregate only, with minimum group sizes and opt-out.

• #1 Answer "What does a typical task cost?" on the pricing page with customers' own published numbers
• #4 One set of usage numbers across the site, from a live counter

• Your websiteEnterprise page: "8M+ minutes of agent runtime" and "290k+ agents run"; Replicas cloud, dedicated single-tenant or self-hosted deployment; SCIM, audit log, DPA and HIPAA BAA; SOC 2 "Type I" with "Type II in progress"; the pricing FAQ has no visible answer.Source
• Your websiteKnowunity story: an AI learning platform for "more than 30 million students"; 30+ engineers on Replicas in July; a target of 25% of PRs and MRs from agents in month one, now 50%; "2×" engineering velocity; bug-fixing time down two-thirds; "2.8k" cloud agents and "400,000+ minutes" in July on 4 vCPU, 16 GB workspaces; daily analytics of usage, compute, token spend and PR output by source and engineer.Source
• Your websiteSitemap (120 URLs, checked November 10, 2026): 61 resource pages, including 16 "alternatives" pages (among them Conductor, Devin, Factory and Codex), 8 "vs" pages and pricing explainers for three rival tools; 28 blog posts, including guides on using PostHog, ClickHouse, Mercury, Vanta and Context Dev with Replicas; 5 customer stories; 5 integration pages; /yc and /trading pages; no changelog, templates, referral or partner page.Source

List the Team and Enterprise plans on the marketplace of the cloud most customers already use for model credentials, with private offers for enterprise deals and the self-hosted option packaged as a deployable product. Co-sell with the provider's startup and partner programmes, and add "buy through your cloud marketplace" to the Enterprise page.

Marketplace billing turns an unbudgeted new vendor into spend against an existing commitment, which shortens enterprise deals for a young company.

• Distraction before product-market fit in enterprise

  Only start after two prospects confirm they would buy this way.

• #14 A security review pack that closes the SOC 2 Type II gap rivals point to

• Your websiteEnterprise page: "8M+ minutes of agent runtime" and "290k+ agents run"; Replicas cloud, dedicated single-tenant or self-hosted deployment; SCIM, audit log, DPA and HIPAA BAA; SOC 2 "Type I" with "Type II in progress"; the pricing FAQ has no visible answer.Source
• Your websiteDocs quickstart ("about 5 minutes"): connect a coding agent subscription or API key with the CLI, connect GitHub or GitLab, add repositories (an environment is created for each), optionally connect Slack or Linear, then "describe a small task" to start the first workspace.Source

Pilot a "Replicas for the whole company" package with two existing customers: ops, sales and support staff request scripts, data pulls and internal tool changes in Slack, an engineer reviews the PR. Measure requests per non-engineer, then write the case study and a landing page for operations leaders.

Every non-engineer request is usage that engineering does not have to schedule, which grows minutes per account without new logos.

• Low-quality requests flood engineers

  Use custom roles and a weekly review cap.

• Your websiteCustomers page: five stories (Knowunity, Parrot, Moda, Utility Profit "ships code across all departments", Vela); only two headlines carry a number (Knowunity "doubled engineering velocity", Vela "3×" faster reviews); a logo wall of 16 companies including Mintlify, Composio and Tensorlake; no quotes on the index.Source
• Your websiteBlog "How We Run Our Software Factory" (October 22, 2026): a #triage Slack channel where an agent turns bugs into Linear tickets; crons for Postgres query tuning, Linear cleanup, a "weekly changelog", stale PR cleanup and a code simplification audit; an agent runs the CRM; the team reviews video or screenshot proof instead of diffs; an autofix agent makes CI green before the PR opens.Source