2026-08-12 · Edu Tor
How the Twin Score rubric actually works
Every human and agent dimension explained with catalog examples, plus category means you can recompute.
Twin Scores are two editorial numbers from 0 to 100, published with weights, not a five-star widget with a made-up rater count. This essay walks through every human dimension and every agent dimension with examples from the 2026 catalog so a reader can falsify us. If a score cannot be explained in public, it does not belong on this directory.
The human axis
Interface clarity asks whether a new user can find the primary action without a training course. Linear and Figma score well because the first session maps onto a craft people already have: issues, frames. Jira and Salesforce score lower because the information architecture is a career.
Onboarding is time-to-value with honest empty states. Clerk and Cal.com get people to a working login or booking page quickly. AWS onboarding is a maze of IAM, which is why its human Twin Score trails its agent score.
Documentation for humans is the tutorial, the screenshot, the conceptual guide. Stripe is the gold standard: copy-paste that works. GA4's help center is large and still leaves marketers unsure which report is real.
Collaboration covers permissions, comments, multiplayer. Figma's multiplayer is the existence proof. Superhuman is almost single-player email speed, so collaboration is not the point — we do not punish a product for a job it refused.
Support, pricing fairness, accessibility, and reliability close the human axis. We record quote-only enterprise pricing as a fairness problem. We do not invent SLA percentages we did not measure.
Worked example: design tools
Midjourney, Canva, Framer, and Figma sit in one category and split on this axis. Canva's human score is high because the job (a deck, a post, a poster) completes. Midjourney's aesthetic range is extraordinary and still Discord-shaped. Figma is the professional canvas with file APIs. Framer ships pretty sites. Treating them as one "best design tool 2026" list is how affiliate pages lie.
The agent axis
API quality is completeness, versioning, idempotency, and error shape. Stripe, GitHub, and the OpenAI Platform are excellent. Consumer Character.AI is none, and we print none.
MCP or tool surface is where 2026 marketing outruns reality. We mark none unless a documented tool surface exists. Cursor and Claude get more than none because they actually expose agent-shaped seams. Most of the catalog is none. That emptiness is the finding.
CLI, structured docs, machine auth, quotas, determinism, and skill packaging finish the axis. A dashboard that exports CSV is not a CLI. A blog post is not structured docs. A cookie session is not machine auth.
Worked example: payments versus painting
Stripe's agent score is 96 because a machine can take a payment. Midjourney's agent score is 42 because a machine cannot honestly complete the same job the GUI does. Both can be beloved. Only one is a machine product.
Category means in this catalog
| Category | Count | Mean human | Mean agent | |---|---:|---:|---:| | ai-assistants | 7 | 81 | 76 | | automation | 3 | 84 | 79 | | communications | 6 | 82 | 80 | | data | 8 | 77 | 86 | | design | 9 | 85 | 66 | | developer-tools | 11 | 83 | 85 | | infrastructure | 13 | 81 | 88 | | marketing | 5 | 77 | 83 | | payments | 1 | 84 | 96 | | productivity | 9 | 83 | 71 |
These means are original research from the committed JSON, not a vendor slide. Design shows the largest human advantage. Infrastructure shows the largest agent advantage. Developer tools cancel out because editors (human) and LLM APIs (agent) share the bucket.
How to challenge a score
Email editor@buildsomethingpeoplewant.com with a public URL that contradicts a capability cell. Named editors update the JSON. We would rather be wrong in public than vague in private. The people who need this page are buyers, journalists, and agents who must not hallucinate an MCP server.
Related
Read the scorecard, the editorial policy, and the 2026 surfaces table. Download the CSV from the agents twin. Remix under CC BY 4.0 and say where it came from.
## Catalog notes (how-twin-score-rubric-works)
ChatGPT (OpenAI) sits at human 88 and agent 90 with API excellent and MCP basic. In the how-twin-score-rubric-works frame: founded 2022 in San Francisco, USA; pricing as publicly stated is Free tier; Plus, Pro, and Business plans. API billed separately on platform.openai.com. Official site https://chatgpt.com.
Claude (Anthropic) sits at human 86 and agent 92 with API excellent and MCP strong. In the how-twin-score-rubric-works frame: founded 2023 in San Francisco, USA; pricing as publicly stated is Free, Pro, Team, and Enterprise on claude.ai; Anthropic API is usage-based. Official site https://claude.ai.
Gemini (Google) sits at human 82 and agent 88 with API excellent and MCP basic. In the how-twin-score-rubric-works frame: founded 2023 in Mountain View, USA; pricing as publicly stated is Free consumer app; Google AI Plus/Ultra; Gemini API and Vertex AI billed in Google Cloud. Official site https://gemini.google.com.
Grok (xAI) sits at human 74 and agent 70 with API strong and MCP none. In the how-twin-score-rubric-works frame: founded 2023 in San Francisco, USA; pricing as publicly stated is Consumer access bundled with X Premium tiers; API available from xAI. Official site https://x.ai.
Perplexity (Perplexity AI) sits at human 85 and agent 78 with API strong and MCP basic. In the how-twin-score-rubric-works frame: founded 2022 in San Francisco, USA; pricing as publicly stated is Free search; Perplexity Pro and Enterprise with higher limits and model choice. Official site https://www.perplexity.ai.
Microsoft Copilot (Microsoft) sits at human 80 and agent 76 with API strong and MCP none. In the how-twin-score-rubric-works frame: founded 2023 in Redmond, USA; pricing as publicly stated is Consumer Copilot free and Pro; Microsoft 365 Copilot is per-user enterprise licensing. Official site https://copilot.microsoft.com.
Midjourney (Midjourney) sits at human 84 and agent 42 with API basic and MCP none. In the how-twin-score-rubric-works frame: founded 2022 in San Francisco, USA; pricing as publicly stated is Subscription GPU time in Basic, Standard, Pro, and Mega monthly plans. Official site https://www.midjourney.com.
Runway (Runway ML) sits at human 81 and agent 64 with API strong and MCP none. In the how-twin-score-rubric-works frame: founded 2018 in New York, USA; pricing as publicly stated is Free trial credits; Standard, Pro, and Unlimited plans with credit or seat billing. Official site https://runwayml.com.
ElevenLabs (ElevenLabs) sits at human 83 and agent 88 with API excellent and MCP basic. In the how-twin-score-rubric-works frame: founded 2022 in London, UK / New York, USA; pricing as publicly stated is Free character cap; Starter through Scale subscriptions; usage-based API. Official site https://elevenlabs.io.
Character.AI (Character.AI) sits at human 72 and agent 35 with API none and MCP none. In the how-twin-score-rubric-works frame: founded 2021 in Menlo Park, USA; pricing as publicly stated is Free with cAI+ subscription for faster responses and features. Official site https://character.ai.
Cursor (Anysphere) sits at human 90 and agent 84 with API strong and MCP strong. In the how-twin-score-rubric-works frame: founded 2023 in San Francisco, USA; pricing as publicly stated is Hobby free; Pro and Business subscription with usage-based model calls. Official site https://cursor.com.
GitHub (Microsoft) sits at human 91 and agent 94 with API excellent and MCP basic. In the how-twin-score-rubric-works frame: founded 2008 in San Francisco, USA; pricing as publicly stated is Free public repos; Team and Enterprise per-seat plans; Actions minutes extra. Official site https://github.com.
GitHub Copilot (GitHub) sits at human 84 and agent 72 with API basic and MCP none. In the how-twin-score-rubric-works frame: founded 2021 in San Francisco, USA; pricing as publicly stated is Individual, Business, and Enterprise seat plans; some features bundled into GitHub editions. Official site https://github.com/features/copilot.
Visual Studio Code (Microsoft) sits at human 93 and agent 80 with API strong and MCP basic. In the how-twin-score-rubric-works frame: founded 2015 in Redmond, USA; pricing as publicly stated is Free and open source (MIT) editor; optional Microsoft account for Settings Sync. Official site https://code.visualstudio.com.
GitLab (GitLab) sits at human 82 and agent 90 with API excellent and MCP none. In the how-twin-score-rubric-works frame: founded 2011 in San Francisco, USA / remote; pricing as publicly stated is Free SaaS tier; Premium and Ultimate; self-managed licenses. Official site https://gitlab.com.
Vercel (Vercel) sits at human 88 and agent 87 with API excellent and MCP basic. In the how-twin-score-rubric-works frame: founded 2015 in San Francisco, USA; pricing as publicly stated is Hobby free; Pro per-seat; Enterprise. Usage on builds, bandwidth, and compute. Official site https://vercel.com.
Netlify (Netlify) sits at human 80 and agent 82 with API strong and MCP none. In the how-twin-score-rubric-works frame: founded 2014 in San Francisco, USA; pricing as publicly stated is Free starter; Pro, Partner, and Enterprise with usage on compute and bandwidth. Official site https://www.netlify.com.
Linear (Linear) sits at human 92 and agent 86 with API excellent and MCP basic. In the how-twin-score-rubric-works frame: founded 2019 in San Francisco, USA / remote; pricing as publicly stated is Free for small teams; Standard, Plus, and Enterprise per-seat. Official site https://linear.app.
Jira (Atlassian) sits at human 61 and agent 84 with API excellent and MCP none. In the how-twin-score-rubric-works frame: founded 2002 in Sydney, Australia; pricing as publicly stated is Free small teams; Standard, Premium, Enterprise. Cloud or Data Center. Official site https://www.atlassian.com/software/jira.
Replit (Replit) sits at human 83 and agent 75 with API strong and MCP basic. In the how-twin-score-rubric-works frame: founded 2016 in Foster City, USA; pricing as publicly stated is Free starter; Core and Teams subscriptions; usage on compute. Official site https://replit.com.
Notion (Notion Labs) sits at human 87 and agent 83 with API excellent and MCP basic. In the how-twin-score-rubric-works frame: founded 2016 in San Francisco, USA; pricing as publicly stated is Free; Plus, Business, Enterprise per-seat. AI add-on historically bundled or metered. Official site https://www.notion.so.
Coda (Coda) sits at human 78 and agent 74 with API strong and MCP none. In the how-twin-score-rubric-works frame: founded 2014 in San Francisco, USA; pricing as publicly stated is Free; Pro, Team, Enterprise. Doc size and automations gated by plan. Official site https://coda.io.
Airtable (Airtable) sits at human 84 and agent 88 with API excellent and MCP none. In the how-twin-score-rubric-works frame: founded 2012 in San Francisco, USA; pricing as publicly stated is Free; Team, Business, Enterprise Cloud. Automation and record limits by plan. Official site https://airtable.com.
Figma (Figma) sits at human 94 and agent 77 with API strong and MCP none. In the how-twin-score-rubric-works frame: founded 2012 in San Francisco, USA; pricing as publicly stated is Free starter; Professional, Organization, Enterprise. FigJam and Dev Mode in the packaging. Official site https://www.figma.com.
Framer (Framer) sits at human 86 and agent 58 with API basic and MCP none. In the how-twin-score-rubric-works frame: founded 2013 in Amsterdam, Netherlands; pricing as publicly stated is Free canvas; site hosting on Mini through Business with CMS and localization add-ons. Official site https://www.framer.com.
Webflow (Webflow) sits at human 85 and agent 72 with API strong and MCP none. In the how-twin-score-rubric-works frame: founded 2013 in San Francisco, USA; pricing as publicly stated is Workspace and site plans; CMS, hosting, and localization billed on the site plan. Official site https://webflow.com.
Canva (Canva) sits at human 89 and agent 63 with API basic and MCP none. In the how-twin-score-rubric-works frame: founded 2012 in Sydney, Australia; pricing as publicly stated is Free; Pro, Teams, Enterprise. Brand kits and Magic Studio on paid plans. Official site https://www.canva.com.
Adobe Creative Cloud (Adobe) sits at human 83 and agent 60 with API basic and MCP none. In the how-twin-score-rubric-works frame: founded 2013 in San Jose, USA; pricing as publicly stated is Photography, Single App, and All Apps subscriptions; enterprise licensing. Official site https://www.adobe.com/creativecloud.html.
Adobe Firefly (Adobe) sits at human 79 and agent 70 with API strong and MCP none. In the how-twin-score-rubric-works frame: founded 2023 in San Jose, USA; pricing as publicly stated is Generative credits included with Creative Cloud plans; extra credits and API usage. Official site https://www.adobe.com/products/firefly.html.
Superhuman (Superhuman) sits at human 88 and agent 40 with API none and MCP none. In the how-twin-score-rubric-works frame: founded 2014 in San Francisco, USA; pricing as publicly stated is Premium per-seat subscription, historically among the most expensive mail clients. Official site https://superhuman.com.
Slack (Salesforce) sits at human 86 and agent 90 with API excellent and MCP basic. In the how-twin-score-rubric-works frame: founded 2013 in San Francisco, USA; pricing as publicly stated is Free; Pro, Business+, Enterprise Grid per-seat. Connectors and AI extras on higher plans. Official site https://slack.com.
Microsoft Teams (Microsoft) sits at human 73 and agent 82 with API excellent and MCP none. In the how-twin-score-rubric-works frame: founded 2017 in Redmond, USA; pricing as publicly stated is Included with many Microsoft 365 plans; additional calling and webinar SKUs. Official site https://www.microsoft.com/microsoft-teams.
Zoom (Zoom) sits at human 84 and agent 80 with API excellent and MCP none. In the how-twin-score-rubric-works frame: founded 2011 in San Jose, USA; pricing as publicly stated is Free basic meetings; Pro, Business, Enterprise with webinar and phone add-ons. Official site https://zoom.us.
Loom (Atlassian) sits at human 87 and agent 55 with API basic and MCP none. In the how-twin-score-rubric-works frame: founded 2015 in San Francisco, USA; pricing as publicly stated is Free; Business and Enterprise. Atlassian bundling varies by SKU. Official site https://www.loom.com.
Discord (Discord) sits at human 85 and agent 88 with API excellent and MCP none. In the how-twin-score-rubric-works frame: founded 2015 in San Francisco, USA; pricing as publicly stated is Free; Nitro consumer; Server Boosts; Discord for Business experiments. Official site https://discord.com.
Cal.com (Cal.com) sits at human 82 and agent 85 with API excellent and MCP basic. In the how-twin-score-rubric-works frame: founded 2021 in San Francisco, USA / remote; pricing as publicly stated is Open-source self-host; Cloud free through Enterprise. Platform/API billed for embed use. Official site https://cal.com.
Calendly (Calendly) sits at human 84 and agent 70 with API strong and MCP none. In the how-twin-score-rubric-works frame: founded 2013 in Atlanta, USA; pricing as publicly stated is Free basic; Standard, Teams, Enterprise. Official site https://calendly.com.
Grammarly (Grammarly) sits at human 86 and agent 48 with API basic and MCP none. In the how-twin-score-rubric-works frame: founded 2009 in San Francisco, USA; pricing as publicly stated is Free; Pro; Business and Enterprise. Generative features on paid plans. Official site https://www.grammarly.com.
Stripe (Stripe) sits at human 84 and agent 96 with API excellent and MCP basic. In the how-twin-score-rubric-works frame: founded 2010 in South San Francisco, USA / Dublin, Ireland; pricing as publicly stated is Pay-as-you-go processing plus billed products (Billing, Tax, Radar, Connect). Official site https://stripe.com.
Supabase (Supabase) sits at human 86 and agent 93 with API excellent and MCP basic. In the how-twin-score-rubric-works frame: founded 2020 in Singapore / remote; pricing as publicly stated is Free; Pro, Team, Enterprise. Compute, storage, and egress metered. Official site https://supabase.com.
Neon (Neon) sits at human 81 and agent 92 with API excellent and MCP basic. In the how-twin-score-rubric-works frame: founded 2021 in San Francisco, USA / Berlin, Germany; pricing as publicly stated is Free; Launch, Scale, Business. Compute and storage separated. Official site https://neon.tech.
PlanetScale (PlanetScale) sits at human 78 and agent 90 with API excellent and MCP none. In the how-twin-score-rubric-works frame: founded 2018 in San Francisco, USA; pricing as publicly stated is Hobby through scalable paid plans; Vitess-powered clustering historically, with Postgres offerings evolving. Official site https://planetscale.com.
Cloudflare (Cloudflare) sits at human 80 and agent 93 with API excellent and MCP basic. In the how-twin-score-rubric-works frame: founded 2010 in San Francisco, USA; pricing as publicly stated is Free CDN/DNS; Pro, Business, Enterprise; Workers and R2 usage-based. Official site https://www.cloudflare.com.
Amazon Web Services (Amazon) sits at human 64 and agent 91 with API excellent and MCP none. In the how-twin-score-rubric-works frame: founded 2006 in Seattle, USA; pricing as publicly stated is Pay-as-you-go across hundreds of services; Enterprise Discount Programs. Official site https://aws.amazon.com.
Resend (Resend) sits at human 85 and agent 94 with API excellent and MCP none. In the how-twin-score-rubric-works frame: founded 2023 in San Francisco, USA; pricing as publicly stated is Free developer tier; Pro and usage on email volume. Official site https://resend.com.
Auth0 (Okta) sits at human 76 and agent 90 with API excellent and MCP none. In the how-twin-score-rubric-works frame: founded 2013 in Bellevue, USA / San Francisco, USA; pricing as publicly stated is Free; B2C/B2B plans with MAU metering; Enterprise. Official site https://auth0.com.
Clerk (Clerk) sits at human 88 and agent 86 with API excellent and MCP none. In the how-twin-score-rubric-works frame: founded 2019 in United States / remote; pricing as publicly stated is Free; Pro and Enterprise with MAU and feature gates. Official site https://clerk.com.
Railway (Railway) sits at human 84 and agent 82 with API strong and MCP none. In the how-twin-score-rubric-works frame: founded 2020 in United States / remote; pricing as publicly stated is Usage-based with credits; Hobby and Pro; Team. Official site https://railway.app.
Render (Render) sits at human 83 and agent 80 with API strong and MCP none. In the how-twin-score-rubric-works frame: founded 2018 in San Francisco, USA; pricing as publicly stated is Free instances with limits; paid services and teams; usage on compute. Official site https://render.com.
Firebase (Google) sits at human 82 and agent 85 with API excellent and MCP none. In the how-twin-score-rubric-works frame: founded 2011 in San Francisco, USA / Mountain View, USA; pricing as publicly stated is Spark free; Blaze pay-as-you-go on Google Cloud. Official site https://firebase.google.com.
OpenAI Platform (OpenAI) sits at human 74 and agent 95 with API excellent and MCP basic. In the how-twin-score-rubric-works frame: founded 2020 in San Francisco, USA; pricing as publicly stated is Usage-based tokens, tools, storage, and fine-tuning. Separate from ChatGPT subscriptions. Official site https://platform.openai.com.
Anthropic API (Anthropic) sits at human 72 and agent 94 with API excellent and MCP strong. In the how-twin-score-rubric-works frame: founded 2021 in San Francisco, USA; pricing as publicly stated is Usage-based tokens; batch discounts; Bedrock and Vertex as alternate bills. Official site https://docs.anthropic.com.
Pinecone (Pinecone) sits at human 70 and agent 90 with API excellent and MCP none. In the how-twin-score-rubric-works frame: founded 2019 in New York, USA; pricing as publicly stated is Serverless and pod-based usage; free starter. Official site https://www.pinecone.io.
Zapier (Zapier) sits at human 88 and agent 75 with API strong and MCP basic. In the how-twin-score-rubric-works frame: founded 2011 in United States / remote; pricing as publicly stated is Free; Professional, Team, Company with task-based billing. Official site https://zapier.com.
Make (Make) sits at human 83 and agent 74 with API strong and MCP none. In the how-twin-score-rubric-works frame: founded 2012 in Prague, Czechia; pricing as publicly stated is Free; Core, Pro, Teams, Enterprise with operation-based billing. Official site https://www.make.com.
n8n (n8n) sits at human 80 and agent 88 with API excellent and MCP basic. In the how-twin-score-rubric-works frame: founded 2019 in Berlin, Germany; pricing as publicly stated is Fair-code self-host; Cloud plans; source-available license (not OSI MIT). Official site https://n8n.io.
HubSpot (HubSpot) sits at human 81 and agent 84 with API excellent and MCP none. In the how-twin-score-rubric-works frame: founded 2006 in Cambridge, USA; pricing as publicly stated is Free CRM; Marketing, Sales, Service, CMS, and Operations Hubs as paid seats/add-ons. Official site https://www.hubspot.com.
Salesforce (Salesforce) sits at human 58 and agent 86 with API excellent and MCP none. In the how-twin-score-rubric-works frame: founded 1999 in San Francisco, USA; pricing as publicly stated is Edition-based seats (Essentials through Unlimited) plus add-on clouds. Quote-driven. Official site https://www.salesforce.com.
Shopify (Shopify) sits at human 88 and agent 89 with API excellent and MCP none. In the how-twin-score-rubric-works frame: founded 2006 in Ottawa, Canada; pricing as publicly stated is Basic, Shopify, Advanced, Plus. App store and payments extra. Official site https://www.shopify.com.
WooCommerce (Automattic) sits at human 74 and agent 78 with API strong and MCP none. In the how-twin-score-rubric-works frame: founded 2011 in San Francisco, USA / worldwide; pricing as publicly stated is Free plugin; hosting, extensions, and Woo payments extra. Official site https://woocommerce.com.
PostHog (PostHog) sits at human 84 and agent 90 with API excellent and MCP none. In the how-twin-score-rubric-works frame: founded 2020 in San Francisco, USA / London, UK / remote; pricing as publicly stated is Generous free; usage-based paid; self-host option. Official site https://posthog.com.
Amplitude (Amplitude) sits at human 79 and agent 82 with API strong and MCP none. In the how-twin-score-rubric-works frame: founded 2011 in San Francisco, USA; pricing as publicly stated is Free starter; Growth, Enterprise. Events and seats. Official site https://amplitude.com.
Mixpanel (Mixpanel) sits at human 78 and agent 81 with API strong and MCP none. In the how-twin-score-rubric-works frame: founded 2009 in San Francisco, USA; pricing as publicly stated is Free; Growth, Enterprise with MTU-style packaging historically evolving. Official site https://mixpanel.com.
Segment (Twilio) sits at human 74 and agent 88 with API excellent and MCP none. In the how-twin-score-rubric-works frame: founded 2011 in San Francisco, USA; pricing as publicly stated is Twilio Segment plans; MTUs and warehouse destinations drive cost. Official site https://segment.com.
Datadog (Datadog) sits at human 77 and agent 91 with API excellent and MCP none. In the how-twin-score-rubric-works frame: founded 2010 in New York, USA; pricing as publicly stated is Per-host, per-million-events, and SKU-by-SKU. Easy to surprise-bill. Official site https://www.datadoghq.com.
Sentry (Sentry) sits at human 86 and agent 88 with API excellent and MCP none. In the how-twin-score-rubric-works frame: founded 2012 in San Francisco, USA / remote; pricing as publicly stated is Free developer; Team, Business, Enterprise with event volume. Self-host possible. Official site https://sentry.io.
Intercom (Intercom) sits at human 80 and agent 83 with API excellent and MCP none. In the how-twin-score-rubric-works frame: founded 2011 in Dublin, Ireland / San Francisco, USA; pricing as publicly stated is Seat and add-on based (Fin AI agent, support, marketing). Quote-heavy. Official site https://www.intercom.com.
Asana (Asana) sits at human 80 and agent 78 with API strong and MCP none. In the how-twin-score-rubric-works frame: founded 2008 in San Francisco, USA; pricing as publicly stated is Personal free; Starter, Advanced, Enterprise per-seat. Official site https://asana.com.
monday.com (monday.com) sits at human 82 and agent 76 with API strong and MCP none. In the how-twin-score-rubric-works frame: founded 2012 in Tel Aviv, Israel; pricing as publicly stated is Free; Basic, Standard, Pro, Enterprise per-seat with item limits. Official site https://monday.com.
Postman (Postman) sits at human 87 and agent 80 with API strong and MCP none. In the how-twin-score-rubric-works frame: founded 2014 in San Francisco, USA; pricing as publicly stated is Free; Basic, Professional, Enterprise. API calls and team features gated. Official site https://www.postman.com.
WordPress (WordPress / Automattic ecosystem) sits at human 83 and agent 79 with API strong and MCP none. In the how-twin-score-rubric-works frame: founded 2003 in Worldwide (open source); pricing as publicly stated is Software is GPL; hosting (WordPress.com, WP Engine, etc.) is paid. Official site https://wordpress.org.
Google Analytics 4 (Google) sits at human 68 and agent 80 with API strong and MCP none. In the how-twin-score-rubric-works frame: founded 2020 in Mountain View, USA; pricing as publicly stated is Free 360-less GA4; Analytics 360 is enterprise paid. Official site https://analytics.google.com.
Written by Edu Tor. Updated 2026-08-25.