Humwork growth playbook

Add a tabbed install block under the hero, with one tab per integration on the homepage list (plugin marketplace, MCP config, web connector, IDE config, CLI config). Each tab shows the exact copy-paste snippet from the README and a 'Get your API key' button that opens signup with the client preselected. Link each integration logo to its tab, and add a 'View on GitHub' link. Then measure how many signups make their first consultation within 24 hours, by client.

Developer tools convert on the first command that works. Showing the snippet before signup proves the 60-second claim, cuts the steps between a click and a first consultation, and sends GitHub traffic (and stars) to the repo that builders check for trust.

[ ] One tab per homepage integration
[ ] Exact snippet from the README, with a copy button
[ ] 'Get your API key' button that opens signup with the client preselected
[ ] Integration logos link to their tab
[ ] 'View on GitHub' link
[ ] Event tracking: tab view, copy, signup, first consultation

• Snippets go stale as clients change

  Generate the block from the README so there is one source.

• Visitors copy the command but never get a key

  Show the key step inside each tab, with a one-click signup.

• Your websiteHomepage (read November 10, 2026): headline "On demand Human Experts for AI Agents."; experts matched "in under 30 seconds"; "Plug & Play Integration" through "One MCP server, API call, or plugin" in "60 seconds", but the page shows no install command, config snippet or documentation link; every Get Started button goes to the app signup; domains listed are software engineering, product and business strategy, marketing and copy, design and UX, plus finance, research, operations, HR and compliance.Source
• GitHubhumworkai/humwork-mcp (MIT licence, 1 star, 1 fork, 28 commits): install commands and config snippets for each supported client; tools consult_expert, send_chat_message, get_chat_messages, close_chat (optional 1-5 rating) and rate_chat; hooks and skills teach the agent to escalate on repeated failures, architectural decisions or unfamiliar domains; plugin sign-in lasts 30 days before re-authentication.Source
• Your websiteSite structure (checked November 10, 2026): /pricing, /docs, /sitemap.xml, /robots.txt and /llms.txt all return 404; the public pages are the homepage, /support, /privacy and /terms, with no blog; the support page gives an email and phone line but no FAQ.Source
• Your websiteIntegrations listed on the homepage: Claude Code, Cursor, Codex, Claude Cowork, Lovable, Replit, ChatGPT, Claude, Gemini, OpenClaw, "Any MCP" and "Any API", none linked to a setup guide. The README also covers Windsurf.Source

Publish the remote server to the official MCP Registry with a clear name, description, tool list and auth method, then claim or submit listings on Glama and the other large directories, all pointing to the homepage install block. Use the same one-line description everywhere: human experts your agent can call when it gets stuck. Check each listing renders the tools correctly and track referral signups from each.

Registry and directory listings are free, permanent distribution in the exact context of intent (someone looking for an MCP server), and they feed the client apps and search results that read from them.

• Listings drift out of date

  Make registry updates part of the release checklist.

• #1 Put the 60-second install on the homepage: one copy-paste block for each agent you support

• MCP directoriesThe official MCP Registry returns 0 servers for "humwork", and Glama has no Humwork listing (both checked November 10, 2026).Source
• GitHubhumworkai/humwork-mcp (MIT licence, 1 star, 1 fork, 28 commits): install commands and config snippets for each supported client; tools consult_expert, send_chat_message, get_chat_messages, close_chat (optional 1-5 rating) and rate_chat; hooks and skills teach the agent to escalate on repeated failures, architectural decisions or unfamiliar domains; plugin sign-in lasts 30 days before re-authentication.Source
• HiringFounding Engineer (San Francisco, $90K-$140K plus 2.5%-5% equity): owns the matching engine, including coverage of rare domains; realtime agent-expert sessions with AI-summarized context handoff; the MCP plugin layer of tools, hooks and skills that decides when an agent should escalate and how it describes the problem; the expert mobile app; and the client dashboard, billing, ratings and observability. The post says Humwork launched as an MCP plugin a few months ago and is now scaling.Source

Publish /pricing with the credit price by purchase tier, the billing increment and minimum charge per consultation, the typical cost of a consultation by domain, what happens if a session ends early, and the enterprise path (volume pricing and invoicing). Add a small example: a stuck debugging session, how long it took and what it cost. Link it from the navigation and the install block.

Builders will not buy non-refundable credits for an unknown cost. A clear price removes the biggest silent objection, and an example cost next to the hours an agent can waste looping makes the value obvious.

• Prices are still changing

  Date the page and show current rates, as the Terms already promise rates before each consultation.

• Your websiteSite structure (checked November 10, 2026): /pricing, /docs, /sitemap.xml, /robots.txt and /llms.txt all return 404; the public pages are the homepage, /support, /privacy and /terms, with no blog; the support page gives an email and phone line but no FAQ.Source
• Your websiteTerms (last updated April 8, 2026): "One credit has an approximate value of $1.00 USD"; consultations are pre-billed in fixed increments with a minimum charge, and rates are disclosed before a consultation starts; "All credit purchases are final and non-refundable", and pre-billed time is not refunded if a session ends early; Humwork keeps a service fee and pays experts, who are independent contractors, through Stripe Connect, typically in 2-7 business days.Source
• Competitor researchTendem by Toloka (homepage): "Finished work, on demand"; 10,000+ vetted experts; pay per task, with deliverables, price and ETA shown before work starts; first three tasks at 50% off; typical turnaround 2-24 hours; connects to AI tools through MCP.See the research Source

Put a visible booking control in the section: an embedded calendar plus a plain link and the founders' email as a fallback, so it works even when scripts are blocked. Add three questions (team size, agents used, domains needed) so the call starts prepared. Send an instant confirmation with the install block and pricing, and track bookings from this section.

Enterprise intent is rare at this stage and every lost booking is expensive. A plain, always-visible link plus a calendar removes the risk of a silent failure.

• Low-fit calls take founder time

  Use the three questions to route small teams to self-serve.

• Your websiteEnterprise: the Book Demo link jumps to a section headed "Want Humwork for your team?" ("Book a call and we'll set up your team in a day.", with priority matching, a custom expert pool, volume pricing and invoicing, and dedicated support); in the page text we read twice on November 10, 2026, that section had no button, form, calendar or email link.Source

Swap the net retention figure for a builder metric, such as median time to fix in software engineering. Date the stats row ("last 30 days"), define resolution (the agent completed its task after the consultation, or the user rated it solved) and link a short methods note. Explain that the expert count grew from 1,000+ at launch to 3,000+ today, which is proof of momentum in itself.

Technical buyers trust numbers they can check. Dated, defined stats turn the change from 87% to 83% into a measured figure, and a growing expert count becomes a story about growth rather than an inconsistency.

• An honest number is lower

  Show the trend and the definition; precision builds more trust than a high figure.

• Your websiteHomepage proof: a stats row of "<2 min" average first reply, "83%" resolution rate and "3,000+" verified experts, with no date or method; the Software Engineering card shows "130% Net retention" with no period, cohort or source; no testimonials, case studies or customer video.Source
• PressTestingCatalog (sponsored, April 15, 2026): at launch Humwork cited 1,000+ experts, an 87% resolution rate in beta, first responses under two minutes and 2,858 questions resolved before public launch; it contrasts Humwork with freelance marketplaces such as Upwork and Fiverr, where humans hire humans.Source

Add an earnings block to the expert section and the expert signup: typical earnings per consultation and per active hour by domain, payout speed, how the matching queue works, and what vetting involves. Add two short, permissioned profiles of active experts (role and domain, no client details). Target the domains where matching is thinnest.

Supply quality decides resolution rate and match speed. Concrete pay and payout speed attract the senior specialists the product promises, especially in rare domains.

• Figures set expectations the queue can't meet

  Show ranges for active experts and say how demand varies by domain.

• Your websiteExpert side: "Get paid for your expertise. AI agents are hiring." and "Earn real money helping AI agents solve hard problems", with no pay figures; Apply as Expert goes to the expert signup.Source
• Your websiteTerms (last updated April 8, 2026): "One credit has an approximate value of $1.00 USD"; consultations are pre-billed in fixed increments with a minimum charge, and rates are disclosed before a consultation starts; "All credit purchases are final and non-refundable", and pre-billed time is not refunded if a session ends early; Humwork keeps a service fee and pays experts, who are independent contractors, through Stripe Connect, typically in 2-7 business days.Source

Publish /docs from the README: install per client, the five tools and their inputs, when to escalate, what context is shared and redacted, billing per consultation and rating. Add /llms.txt that summarises the product and links to the docs and pricing, plus a sitemap and robots file. Write it so an agent can follow it without a human.

When agents choose and install tools, the vendor whose docs the agent can parse wins. It also helps AI search answers describe Humwork accurately.

• Docs and README drift apart

  Generate both from one source.

• #1 Put the 60-second install on the homepage: one copy-paste block for each agent you support

• Your websiteSite structure (checked November 10, 2026): /pricing, /docs, /sitemap.xml, /robots.txt and /llms.txt all return 404; the public pages are the homepage, /support, /privacy and /terms, with no blog; the support page gives an email and phone line but no FAQ.Source
• GitHubhumworkai/humwork-mcp (MIT licence, 1 star, 1 fork, 28 commits): install commands and config snippets for each supported client; tools consult_expert, send_chat_message, get_chat_messages, close_chat (optional 1-5 rating) and rate_chat; hooks and skills teach the agent to escalate on repeated failures, architectural decisions or unfamiliar domains; plugin sign-in lasts 30 days before re-authentication.Source

Give every new account one free consultation (or a starter credit sized to the minimum charge), say so in the hero and on the pricing page, and trigger it only when the agent actually escalates. After the free session, show what it would have cost and offer a small first pack.

The value of Humwork is felt in the moment an agent stops looping. A free first rescue gets builders to that moment without the risk of buying non-refundable credits up front.

• Free sessions cost real expert time

  Cap at the minimum charge and require a verified account.

• #3 Publish a pricing page: what a credit buys, the minimum charge and a typical consultation

• Your websiteTerms (last updated April 8, 2026): "One credit has an approximate value of $1.00 USD"; consultations are pre-billed in fixed increments with a minimum charge, and rates are disclosed before a consultation starts; "All credit purchases are final and non-refundable", and pre-billed time is not refunded if a session ends early; Humwork keeps a service fee and pays experts, who are independent contractors, through Stripe Connect, typically in 2-7 business days.Source
• Directory listingEveryDev lists Humwork as freemium with a free tier, and shows no ratings or reviews.Source
• Competitor researchTendem by Toloka (homepage): "Finished work, on demand"; 10,000+ vetted experts; pay per task, with deliverables, price and ETA shown before work starts; first three tasks at 50% off; typical turnaround 2-24 hours; connects to AI tools through MCP.See the research Source

Log every escalation with the attempts before it, the time spent looping and the outcome. Then test trigger rules (for example, after the same error twice, or before a schema or architecture change) and let users set a budget and sensitivity. Improve the problem summary the agent sends, since a better brief shortens consultations. Report resolution and cost by trigger type.

Well-timed escalations raise resolution rates and lower cost per fix, which keeps users from switching the plugin off. This is product-led growth sitting inside someone else's product.

• Aggressive triggers feel like upselling

  Default to conservative, with a visible budget.

• GitHubhumworkai/humwork-mcp (MIT licence, 1 star, 1 fork, 28 commits): install commands and config snippets for each supported client; tools consult_expert, send_chat_message, get_chat_messages, close_chat (optional 1-5 rating) and rate_chat; hooks and skills teach the agent to escalate on repeated failures, architectural decisions or unfamiliar domains; plugin sign-in lasts 30 days before re-authentication.Source
• HiringFounding Engineer (San Francisco, $90K-$140K plus 2.5%-5% equity): owns the matching engine, including coverage of rare domains; realtime agent-expert sessions with AI-summarized context handoff; the MCP plugin layer of tools, hooks and skills that decides when an agent should escalate and how it describes the problem; the expert mobile app; and the client dashboard, billing, ratings and observability. The post says Humwork launched as an MCP plugin a few months ago and is now scaling.Source

For each integration on the homepage, find its official directory or marketplace and submit Humwork with a one-click install where supported. Prioritise by active users today. Give each tool a landing page (/integrations/<tool>) with its install steps and a short demo, and link the directory listing to it.

In-product directories put Humwork one click from the user, inside the client they already run, with the client's trust attached.

• Review queues are slow

  Submit to several at once and lead with the fastest.

• #1 Put the 60-second install on the homepage: one copy-paste block for each agent you support
• #2 List Humwork in the official MCP Registry and the main MCP directories

• Your websiteIntegrations listed on the homepage: Claude Code, Cursor, Codex, Claude Cowork, Lovable, Replit, ChatGPT, Claude, Gemini, OpenClaw, "Any MCP" and "Any API", none linked to a setup guide. The README also covers Windsurf.Source
• GitHubhumworkai/humwork-mcp (MIT licence, 1 star, 1 fork, 28 commits): install commands and config snippets for each supported client; tools consult_expert, send_chat_message, get_chat_messages, close_chat (optional 1-5 rating) and rate_chat; hooks and skills teach the agent to escalate on repeated failures, architectural decisions or unfamiliar domains; plugin sign-in lasts 30 days before re-authentication.Source

When close_chat runs, send the user (email and dashboard) a short card: the problem, what the expert changed or advised, time to fix, credits spent, the expert's domain and a one-click rating. Add a monthly digest: consultations, fixes and estimated hours saved. Let users share a card publicly with sensitive details removed.

People keep paying for what they can see working. A visible receipt of each rescue builds the habit, gives users a story to tell colleagues, and gives Humwork real ratings from humans.

• Summaries expose sensitive code

  Use the same redaction as the handoff and keep cards private by default.

• GitHubhumworkai/humwork-mcp (MIT licence, 1 star, 1 fork, 28 commits): install commands and config snippets for each supported client; tools consult_expert, send_chat_message, get_chat_messages, close_chat (optional 1-5 rating) and rate_chat; hooks and skills teach the agent to escalate on repeated failures, architectural decisions or unfamiliar domains; plugin sign-in lasts 30 days before re-authentication.Source
• Y CombinatorSpring 2026 batch, founded 2025, a team of 2 in San Francisco; the launch post (about 25 upvotes) asks AI agent companies and internal agent teams to integrate Humwork as an escalation layer, and experts in engineering, law, marketing, design and medicine to join; a demo video is on YouTube.Source

Track token expiry per account. Email and show a dashboard notice a few days before expiry with a one-click renewal link, refresh tokens silently where the client allows it, and make sure an expired call returns a clear message the agent passes on ('reconnect Humwork: <link>'). Report how many accounts are lapsed each week.

Escalations are rare but valuable. If the tool is disconnected at the moment of need, the user decides it doesn't work. Keeping the connection alive protects every future consultation.

• Longer sessions raise security concerns

  Prefer reminders and refresh tokens over longer-lived credentials.

• GitHubhumworkai/humwork-mcp (MIT licence, 1 star, 1 fork, 28 commits): install commands and config snippets for each supported client; tools consult_expert, send_chat_message, get_chat_messages, close_chat (optional 1-5 rating) and rate_chat; hooks and skills teach the agent to escalate on repeated failures, architectural decisions or unfamiliar domains; plugin sign-in lasts 30 days before re-authentication.Source

Publish a page per domain: the typical stuck moments in that domain, sample expert profiles (role and seniority), median time to fix, cost, and an install path for the agents that domain uses. Start with the two non-engineering domains that already get the most consultations. Link them from the homepage domain cards.

Domain pages rank for specific intent and let non-developer agent users see themselves in the product, which widens the market beyond coding agents.

• Thin supply in a featured domain

  Only publish domains with reliable match times.

• Your websiteHomepage (read November 10, 2026): headline "On demand Human Experts for AI Agents."; experts matched "in under 30 seconds"; "Plug & Play Integration" through "One MCP server, API call, or plugin" in "60 seconds", but the page shows no install command, config snippet or documentation link; every Get Started button goes to the app signup; domains listed are software engineering, product and business strategy, marketing and copy, design and UX, plus finance, research, operations, HR and compliance.Source
• Your websiteSite structure (checked November 10, 2026): /pricing, /docs, /sitemap.xml, /robots.txt and /llms.txt all return 404; the public pages are the homepage, /support, /privacy and /terms, with no blog; the support page gives an email and phone line but no FAQ.Source
• Y CombinatorSpring 2026 batch, founded 2025, a team of 2 in San Francisco; the launch post (about 25 upvotes) asks AI agent companies and internal agent teams to integrate Humwork as an escalation layer, and experts in engineering, law, marketing, design and medicine to join; a demo video is on YouTube.Source

With permission and redaction, record one real rescue a week as a short screen capture: the loop, the escalation, the expert's fix, the agent finishing, and the time and cost. Post it natively on X and LinkedIn with the install link, and embed the best three on the homepage under the hero.

Developers share demos that show a familiar pain resolved. Real rescues are more believable than claims and feed both social reach and homepage conversion.

• Customer code shown by mistake

  Redact and get sign-off before posting.

• Your websiteHomepage proof: a stats row of "<2 min" average first reply, "83%" resolution rate and "3,000+" verified experts, with no date or method; the Software Engineering card shows "130% Net retention" with no period, cohort or source; no testimonials, case studies or customer video.Source
• Y CombinatorSpring 2026 batch, founded 2025, a team of 2 in San Francisco; the launch post (about 25 upvotes) asks AI agent companies and internal agent teams to integrate Humwork as an escalation layer, and experts in engineering, law, marketing, design and medicine to join; a demo video is on YouTube.Source
• ResearchChannels: the site links only X and LinkedIn; no YouTube channel, newsletter, blog or public status page; no Hacker News mentions found.Source

Add a team workspace: invite teammates, share a credit pool, set per-user and per-agent budgets, see consultations across the team, and pay by card or monthly invoice above a threshold. Keep custom expert pools and priority matching for enterprise. Promote it on the pricing page and in the Book Demo section for small teams.

Team plans raise revenue per account and spread usage across users, which smooths the uneven, need-driven pattern of individual escalations.

• Cannibalises enterprise deals

  Keep custom pools and priority matching enterprise-only.

• #3 Publish a pricing page: what a credit buys, the minimum charge and a typical consultation

• Your websiteEnterprise: the Book Demo link jumps to a section headed "Want Humwork for your team?" ("Book a call and we'll set up your team in a day.", with priority matching, a custom expert pool, volume pricing and invoicing, and dedicated support); in the page text we read twice on November 10, 2026, that section had no button, form, calendar or email link.Source
• Your websiteTerms (last updated April 8, 2026): "One credit has an approximate value of $1.00 USD"; consultations are pre-billed in fixed increments with a minimum charge, and rates are disclosed before a consultation starts; "All credit purchases are final and non-refundable", and pre-billed time is not refunded if a session ends early; Humwork keeps a service fee and pays experts, who are independent contractors, through Stripe Connect, typically in 2-7 business days.Source
• Y CombinatorSpring 2026 batch, founded 2025, a team of 2 in San Francisco; the launch post (about 25 upvotes) asks AI agent companies and internal agent teams to integrate Humwork as an escalation layer, and experts in engineering, law, marketing, design and medicine to join; a demo video is on YouTube.Source

Log every request that took too long to match or went unresolved, by domain and time zone. Each month, publish the top gaps to current experts and pay a referral bonus when a referred expert passes vetting and completes five consultations. Add a 'refer a colleague' link to the expert app and to payout emails.

Specialists know other specialists. Referrals pointed at measured gaps raise match speed and resolution where it matters most, at lower cost than broad recruiting.

• Referred experts are lower quality

  Same vetting for everyone; bonus only after five rated consultations.

• #6 Show experts what they can earn and when they get paid

• HiringFounding Engineer (San Francisco, $90K-$140K plus 2.5%-5% equity): owns the matching engine, including coverage of rare domains; realtime agent-expert sessions with AI-summarized context handoff; the MCP plugin layer of tools, hooks and skills that decides when an agent should escalate and how it describes the problem; the expert mobile app; and the client dashboard, billing, ratings and observability. The post says Humwork launched as an MCP plugin a few months ago and is now scaling.Source
• Your websiteExpert side: "Get paid for your expertise. AI agents are hiring." and "Earn real money helping AI agents solve hard problems", with no pay figures; Apply as Expert goes to the expert signup.Source
• Your websiteHomepage (read November 10, 2026): headline "On demand Human Experts for AI Agents."; experts matched "in under 30 seconds"; "Plug & Play Integration" through "One MCP server, API call, or plugin" in "60 seconds", but the page shows no install command, config snippet or documentation link; every Get Started button goes to the app signup; domains listed are software engineering, product and business strategy, marketing and copy, design and UX, plus finance, research, operations, HR and compliance.Source

With customer consent and redaction, distil resolved consultations into short verified fixes (problem signature, fix, expert domain, date). Add a lookup tool the agent calls before consult_expert, priced at a fraction of a consultation, with a share paid to the expert whose fix is reused. Show 'answered from verified fixes' in the session summary.

It creates a cheaper entry tier, pays experts for reusable work, and builds a data asset competitors with hour-long turnarounds can't match. Each consultation makes the next one cheaper.

• Customers object to reuse of their sessions

  Opt-in only, generalised fixes, no code or names.

• Library cannibalises consultations

  Price for volume; humans stay on new problems.

• #9 Tune when the agent calls for help: the escalation trigger is your real growth lever

• PressTestingCatalog (sponsored, April 15, 2026): at launch Humwork cited 1,000+ experts, an 87% resolution rate in beta, first responses under two minutes and 2,858 questions resolved before public launch; it contrasts Humwork with freelance marketplaces such as Upwork and Fiverr, where humans hire humans.Source
• GitHubhumworkai/humwork-mcp (MIT licence, 1 star, 1 fork, 28 commits): install commands and config snippets for each supported client; tools consult_expert, send_chat_message, get_chat_messages, close_chat (optional 1-5 rating) and rate_chat; hooks and skills teach the agent to escalate on repeated failures, architectural decisions or unfamiliar domains; plugin sign-in lasts 30 days before re-authentication.Source
• Competitor researchTendem by Toloka (homepage): "Finished work, on demand"; 10,000+ vetted experts; pay per task, with deliverables, price and ETA shown before work starts; first three tasks at 50% off; typical turnaround 2-24 hours; connects to AI tools through MCP.See the research Source

Add a review mode: before an agent ships a contract clause, a financial model, a migration or public copy, it sends the output to a vetted expert in that domain, who approves, edits or rejects it. The record (who reviewed, credentials, what changed, when) goes into a downloadable log. Price it per review, with enterprise SLAs. Be clear about what a review is and isn't (for example, not legal advice unless the expert's terms say so).

Teams deploying agents for higher-stakes work need a human on record, and will pay more for it than for debugging help. It also sells to the enterprise buyers the Book Demo section targets.

• Liability for reviewed work

  Take legal advice, define scope in the terms, and start with lower-risk domains.

• Slow reviews break the agent loop

  Async mode with notifications for long reviews.

• Competitor researchgotoHuman (homepage): "AI Automation unlocked with human supervision"; reviewers approve and edit agent actions; integrations with n8n, Make, JavaScript, Python and LangGraph plus an MCP server; a case study reports 40 hours saved monthly.See the research Source
• Competitor researchPearl's MCP server (blog, May 2025): tools for a fast AI answer, an AI intake then a human expert, or a direct expert answer, across legal, veterinary, automotive, insurance, financial and medical fields; API keys requested by email.See the research Source
• Y CombinatorSpring 2026 batch, founded 2025, a team of 2 in San Francisco; the launch post (about 25 upvotes) asks AI agent companies and internal agent teams to integrate Humwork as an escalation layer, and experts in engineering, law, marketing, design and medicine to join; a demo video is on YouTube.Source
• Your websiteEnterprise: the Book Demo link jumps to a section headed "Want Humwork for your team?" ("Book a call and we'll set up your team in a day.", with priority matching, a custom expert pool, volume pricing and invoicing, and dedicated support); in the page text we read twice on November 10, 2026, that section had no button, form, calendar or email link.Source

Package a partner API: white-label or co-branded 'Get a human expert' inside the partner's product, partner-level billing with a revenue share, a partner dashboard, and usage limits per end user. Sign two design partners among agent and app-builder companies, measure their users' completion rates with and without escalation, and publish the result as a case study.

One partner integration can bring thousands of end users at once, without Humwork paying to reach them, and partners gain fewer abandoned projects.

• Partners build their own

  Compete on supply: the vetted network is the hard part.

• Supply can't meet partner spikes

  Per-partner limits and priority pools.

• #15 A self-serve team plan: shared credits, seats and invoices without a sales call

• Y CombinatorSpring 2026 batch, founded 2025, a team of 2 in San Francisco; the launch post (about 25 upvotes) asks AI agent companies and internal agent teams to integrate Humwork as an escalation layer, and experts in engineering, law, marketing, design and medicine to join; a demo video is on YouTube.Source
• HiringFounding Engineer (San Francisco, $90K-$140K plus 2.5%-5% equity): owns the matching engine, including coverage of rare domains; realtime agent-expert sessions with AI-summarized context handoff; the MCP plugin layer of tools, hooks and skills that decides when an agent should escalate and how it describes the problem; the expert mobile app; and the client dashboard, billing, ratings and observability. The post says Humwork launched as an MCP plugin a few months ago and is now scaling.Source
• Your websiteIntegrations listed on the homepage: Claude Code, Cursor, Codex, Claude Cowork, Lovable, Replit, ChatGPT, Claude, Gemini, OpenClaw, "Any MCP" and "Any API", none linked to a setup guide. The README also covers Windsurf.Source

Each month, publish aggregated, anonymised findings: top failure types by domain, median time to fix, which problems humans solve fastest, and trends. Share charts on X and LinkedIn, offer the report by email (building the newsletter Humwork doesn't have), and pitch the findings to developer newsletters.

Original data earns links, press and trust in a new category, and positions Humwork as the authority on human escalation. Each report also reminds readers when to call an expert.

• Data reveals customer details

  Aggregate only, minimum group sizes, no code.

• PressTestingCatalog (sponsored, April 15, 2026): at launch Humwork cited 1,000+ experts, an 87% resolution rate in beta, first responses under two minutes and 2,858 questions resolved before public launch; it contrasts Humwork with freelance marketplaces such as Upwork and Fiverr, where humans hire humans.Source
• ResearchChannels: the site links only X and LinkedIn; no YouTube channel, newsletter, blog or public status page; no Hacker News mentions found.Source

Add a small live widget to the homepage and domain pages: experts online now per domain, median match time in the last hour, and consultations resolved today. Add a public status page for the matching service. Show the same numbers in the dashboard before a user buys credits.

Live numbers turn a claim into evidence and make the marketplace feel busy. They also hold the team to the 30-second promise in public.

• Low numbers at quiet hours

  Show median match time, which stays low, rather than raw counts when supply is thin.

• #5 Replace "130% Net retention" with dated stats a builder can trust

• Your websiteHomepage (read November 10, 2026): headline "On demand Human Experts for AI Agents."; experts matched "in under 30 seconds"; "Plug & Play Integration" through "One MCP server, API call, or plugin" in "60 seconds", but the page shows no install command, config snippet or documentation link; every Get Started button goes to the app signup; domains listed are software engineering, product and business strategy, marketing and copy, design and UX, plus finance, research, operations, HR and compliance.Source
• ResearchChannels: the site links only X and LinkedIn; no YouTube channel, newsletter, blog or public status page; no Hacker News mentions found.Source
• Your websiteHomepage proof: a stats row of "<2 min" average first reply, "83%" resolution rate and "3,000+" verified experts, with no date or method; the Software Engineering card shows "130% Net retention" with no period, cohort or source; no testimonials, case studies or customer video.Source

Serve the mockup and logo images in modern formats at the right sizes, lazy-load below-the-fold images, and inline critical CSS to clear the render-blocking request. Add an Open Graph image showing the agent-to-expert handoff, plus a title and description, and check previews on X and LinkedIn.

A preview card lifts click-through on every social share, and a lighter page keeps the fast experience as content grows.

• Compressed mockups look blurry

  Check on high-density screens before shipping.

• PageSpeedMobile performance 96, desktop 100; real visitors pass Core Web Vitals (LCP p75 1.41 s, INP 49 ms, CLS 0), though field LCP is trending about 10% slower; the homepage weighs 2.2 MB, 1.68 MB of it images, with about 990 KB saveable; no social share image.Source
• ResearchChannels: the site links only X and LinkedIn; no YouTube channel, newsletter, blog or public status page; no Hacker News mentions found.Source