Marblism Research · Technical report · v1.0 · October 7, 2026

AI Employee Index

Can AI tools actually run the back office of a service business? We graded 6 products on 27 real jobs, from answering the phone to sending quotes.

  1. 1Marblism logoMarblism25/27 done · $44/mo*94.4
  2. 2Lindy logoLindy15/27 done · $29.99/mo*66.7
  3. 3Grok logoGrok12/27 done · $30/mo63.0
  4. 4Sintra logoSintra13/27 done · $48.50/mo*61.1
  5. 5ChatGPT logoChatGPT10/27 done · $20/mo59.3
  6. 6Claude logoClaude8/27 done · $20/mo53.7
Figure 1 · AI Employee Index, higher is better. One point for each of 27 jobs a product does on its own, half a point if it only partly does it. Standard self-serve plans, built-in features only. *Monthly billing; see Table 2 for price notes.

Summary

AI benchmarks usually measure how smart a model is. People who run a service business care about something else: whether the work gets done while they're out on a job. The AI Employee Index grades 6 AI products on 27 jobs across 8 categories, using each product's standard self-serve plan, its built-in features only, and its own documentation.

Marblism scores 94.4. The strongest alternative, Lindy, scores 66.7. No other product can answer a business phone line, and every product, ours included, falls short somewhere.

1Key findings

  1. 1.1

    Nobody else picks up the phone. Marblism is the only product that answers a business line, books the job during the call, and transfers to your cell. Every alternative scores 0 on front desk, and for a contractor that's where the money comes in.

  2. 1.2

    Chatbots help, but you still do the work. ChatGPT and Claude only partly do 12 and 13 of the 27 jobs. They'll draft it if you ask, or if you wire it up yourself, but they won't run it on their own.

  3. 1.3

    Lindy matches us on the inbox. Both score 91.7 on inbox & calendar. If email is your only problem, it's a real option.

  4. 1.4

    Where we fall short. Eva never sends an email without you; that's by design. And you can't build your own AI agent with Marblism. Every other product lets you, which is why Grok beats us on teamwork & setup (87.5 vs 75.0).

  5. 1.5

    Paperwork is a solved problem. Every product can draft a quote and review a contract. It's no longer a reason to pick one tool over another.

2Results

Figure 1 ranks the overall index. Below, the same results split by job category, set against price, and laid out job by job.

Figure 2

Score by job category (higher is better)

Front desk

3 jobs

  • Marblism100.0
  • Lindy0.0
  • Grok0.0
  • Sintra0.0
  • ChatGPT0.0
  • Claude0.0

Inbox & calendar

6 jobs

  • Marblism91.7
  • Lindy91.7
  • Grok58.3
  • Sintra75.0
  • ChatGPT75.0
  • Claude50.0

Finding customers

3 jobs

  • Marblism100.0
  • Lindy33.3
  • Grok50.0
  • Sintra16.7
  • ChatGPT50.0
  • Claude50.0

Getting found on Google

3 jobs

  • Marblism100.0
  • Lindy66.7
  • Grok66.7
  • Sintra83.3
  • ChatGPT66.7
  • Claude66.7

Social media

3 jobs

  • Marblism100.0
  • Lindy83.3
  • Grok83.3
  • Sintra66.7
  • ChatGPT66.7
  • Claude66.7

Website

3 jobs

  • Marblism100.0
  • Lindy66.7
  • Grok66.7
  • Sintra66.7
  • ChatGPT50.0
  • Claude50.0

Paperwork

2 jobs

  • Marblism100.0
  • Lindy100.0
  • Grok100.0
  • Sintra100.0
  • ChatGPT100.0
  • Claude100.0

Teamwork & setup

4 jobs

  • Marblism75.0
  • Lindy75.0
  • Grok87.5
  • Sintra75.0
  • ChatGPT62.5
  • Claude62.5
Share of available points in each category. Products appear in overall-rank order in every panel.
Figure 3

Index score vs. monthly price

40
60
80
100
$0$10$20$30$40$50ScorePrice per month →
Marblism 94.4
Lindy 66.7
Grok 63.0
Sintra 61.1
ChatGPT 59.3
Claude 53.7
Prices on monthly billing, checked October 7, 2026. Marblism is $24/mo billed yearly and includes unlimited team members; Lindy is priced per user; Sintra's price is promotional ($97 list).
Table 1

Every job, every product

  • Does it on its own (1 pt)
  • Partly (½ pt)
  • Can't (0 pt)

Scroll sideways to see every product →

JobMarblismLindyGrokSintraChatGPTClaude
Front desk
R1Answers your business phone line 24/7
R2Books appointments during the call
R3Transfers calls to your cell when needed
Inbox & calendar
E1Sorts your inbox and filters junk on its own
E2Drafts replies in your voice, waiting in your inbox
E3Joins your meetings and takes notes
E4Sends you task reminders
E5Works with both Gmail and Outlook
E6Replies to routine emails without you clicking send
Finding customers
L1Finds leads from a built-in database
L2Sends cold emails and follow-ups automatically
L3Does LinkedIn outreach automatically
Getting found on Google
P1Writes SEO blog articles
P2Publishes them to your website on its own
P3Keyword research and SEO audit
Social media
S1Writes posts in your tone
S2Publishes to Instagram, TikTok, YouTube, LinkedIn, Threads, X and Facebook on a schedule
S3Creates images, carousels and videos
Website
W1Builds and hosts your website from a prompt
W2Runs A/B tests on your site around the clock
W3Edits your website when you ask in chat
Paperwork
Q1Drafts quotes, estimates and invoices
G1Reviews contracts and flags risky clauses
Teamwork & setup
T1AI employees hand off work to each other
T2Group chat with your AI and human team
T3Separate workspaces for each business
T4Lets you build your own AI agent for any job
Index score94.466.763.061.159.353.7
Hover a cell for the reasoning. Notes and sources for every verdict are in Appendix A.

3Method

3.1 The jobs

27 jobs in 8 categories that contractors and service businesses want off their plate (see Table 1). Most come from what people hire Marblism for. We added jobs where we know we fall short (sending email without approval, building your own agent) and jobs every tool can do (quotes, contracts), so the test isn't only played on our home turf.

3.2 Products and plans

Each product was scored on the plan a typical owner would buy on their own, using only built-in features. Zapier, Make, n8n, custom code and developer APIs don't count.

Table 2 Products evaluated
ProductPlanMonthlyNotes
MarblismAll AI employees$44$24/mo billed yearly · unlimited team members
LindyPlus$29.99per user
GrokSuperGrok$30—
SintraSintra X$48.50promotional price; $97 list
ChatGPTPlus$20—
ClaudePro$20—

3.3 Grading

Every job gets one of three grades. When two grades were defensible, we gave the other product the higher one. When two products do something the same way, they get the same grade.

Table 3 Grading rubric
PointsGradeMeaningExample
1Does it on its ownA built-in feature or ready-made template does the job. If the job is ongoing (“24/7”, “on a schedule”), it runs without you asking each time.Lindy labels every incoming email on its own.
½PartlyYou have to ask every time, build it yourself, approve each action, or it only covers one platform. Limited betas count here too.Claude can draft Gmail replies, but you set up the scheduled task.
0Can'tNo built-in way to do the job.None of the chatbots has a business phone number.

3.4 The index

Index = points ÷ 27 × 100

Every job weighs the same. If answering the phone matters more to you than posting on Instagram, the raw grades in Table 1 let you weigh them your own way.

4Limitations

  • We wrote it. Marblism designed the job list, and most jobs come from what our customers ask for. Read the results with that in mind. That's why every grade and its source is public.
  • Whether, not how well. We checked whether a product can do a job, not how well it does it.
  • Documentation, not hands-on testing. Grades come from each company's own website and help docs. Where marketing pages and the help center disagreed, we went with the help center.
  • A snapshot. Checked October 7, 2026. These products ship every month; we'll re-run the index and log every change.
  • One plan each. Team and enterprise tiers unlock more. We scored the plan an owner would buy on their own.

Spotted a mistake? Email support@marblism.com and we'll correct it in the next version.

AEvidence

The reasoning and source behind every grade. Marblism's grades rely on what we publish on our AI employee pages.

Marblism · All AI employees · $44/mo
  1. R1Answers your business phone line 24/7. Published on our pricing page and AI employee pages.
  2. R2Books appointments during the call. Published on our pricing page and AI employee pages.
  3. R3Transfers calls to your cell when needed. Published on our pricing page and AI employee pages.
  4. E1Sorts your inbox and filters junk on its own. Published on our pricing page and AI employee pages.
  5. E2Drafts replies in your voice, waiting in your inbox. Published on our pricing page and AI employee pages.
  6. E3Joins your meetings and takes notes. Published on our pricing page and AI employee pages.
  7. E4Sends you task reminders. Published on our pricing page and AI employee pages.
  8. E5Works with both Gmail and Outlook. Published on our pricing page and AI employee pages.
  9. E6Replies to routine emails without you clicking send. By design, Eva prepares drafts in your inbox and a human clicks send.
  10. L1Finds leads from a built-in database. Published on our pricing page and AI employee pages.
  11. L2Sends cold emails and follow-ups automatically. Published on our pricing page and AI employee pages.
  12. L3Does LinkedIn outreach automatically. Published on our pricing page and AI employee pages.
  13. P1Writes SEO blog articles. Published on our pricing page and AI employee pages.
  14. P2Publishes them to your website on its own. Published on our pricing page and AI employee pages.
  15. P3Keyword research and SEO audit. Published on our pricing page and AI employee pages.
  16. S1Writes posts in your tone. Published on our pricing page and AI employee pages.
  17. S2Publishes to Instagram, TikTok, YouTube, LinkedIn, Threads, X and Facebook on a schedule. Published on our pricing page and AI employee pages.
  18. S3Creates images, carousels and videos. Published on our pricing page and AI employee pages.
  19. W1Builds and hosts your website from a prompt. Published on our pricing page and AI employee pages.
  20. W2Runs A/B tests on your site around the clock. Published on our pricing page and AI employee pages.
  21. W3Edits your website when you ask in chat. Published on our pricing page and AI employee pages.
  22. Q1Drafts quotes, estimates and invoices. Published on our pricing page and AI employee pages.
  23. G1Reviews contracts and flags risky clauses. Published on our pricing page and AI employee pages.
  24. T1AI employees hand off work to each other. Published on our pricing page and AI employee pages.
  25. T2Group chat with your AI and human team. Published on our pricing page and AI employee pages.
  26. T3Separate workspaces for each business. Published on our pricing page and AI employee pages.
  27. T4Lets you build your own AI agent for any job. You hire from our team of AI employees; you can't build your own from scratch.
Lindy · Plus · $29.99/mo
  1. R1Answers your business phone line 24/7. The current Lindy (Slack-based Teammate, Aug 2026) has no phone or voice agent on any plan; its number is for texting Lindy. source
  2. R2Books appointments during the call. No voice agent; scheduling works over email, chat and text, not during a call. source
  3. R3Transfers calls to your cell when needed. No inbound call handling, so no transfers.
  4. E1Sorts your inbox and filters junk on its own. Email labeling runs on every incoming email and archives what you don't need, on Gmail or Outlook. source
  5. E2Drafts replies in your voice, waiting in your inbox. Drafts replies in your tone and leaves them in your drafts folder. source
  6. E3Joins your meetings and takes notes. A notetaker joins Meet, Zoom and Teams and produces transcripts and summaries. source
  7. E4Sends you task reminders. One-off or recurring reminders by text, Slack or chat. source
  8. E5Works with both Gmail and Outlook. Inbox features work with both Gmail and Outlook. source
  9. E6Replies to routine emails without you clicking send. Built-in email routines only save drafts; auto-sending means building your own routine. source
  10. L1Finds leads from a built-in database. No built-in contacts database; it researches leads on the web or searches your own Apollo account. source
  11. L2Sends cold emails and follow-ups automatically. Drafts personalized sequences and follow-ups from a lead sheet; sending waits for your approval. source
  12. L3Does LinkedIn outreach automatically. LinkedIn integration can post, comment and like, but has no messaging or connection requests. source
  13. P1Writes SEO blog articles. Writes articles on request with web research. source
  14. P2Publishes them to your website on its own. WordPress and Webflow publishing actions exist, but you set up the routine and publishing needs approval. source
  15. P3Keyword research and SEO audit. Can watch rankings and review a site using web research; no keyword-volume tool. source
  16. S1Writes posts in your tone. Drafts content and keeps your style through memory. source
  17. S2Publishes to Instagram, TikTok, YouTube, LinkedIn, Threads, X and Facebook on a schedule. Can post to LinkedIn on a routine you set up; no Instagram, Facebook or X posting. source
  18. S3Creates images, carousels and videos. Built-in image generation, and a video use case that scripts and cuts videos. source
  19. W1Builds and hosts your website from a prompt. Lindy Build creates sites from a prompt and hosts them on lindy.site links. source
  20. W2Runs A/B tests on your site around the clock. No website A/B testing feature.
  21. W3Edits your website when you ask in chat. Rebuilds the hosted page when you ask for changes in chat. source
  22. Q1Drafts quotes, estimates and invoices. QuickBooks integration creates invoices and estimates; it can also create Stripe invoices. source
  23. G1Reviews contracts and flags risky clauses. No dedicated feature, but it reads a contract you share and comments on risky clauses on request. source
  24. T1AI employees hand off work to each other. The current product is a single AI teammate; no multi-agent handoff is documented. source
  25. T2Group chat with your AI and human team. Works inside Slack channels and threads with your human teammates. source
  26. T3Separate workspaces for each business. Workspace switcher keeps each workspace's files separate. source
  27. T4Lets you build your own AI agent for any job. Create your own skills and routines that act in your tools on a schedule or trigger. source
Grok · SuperGrok · $30/mo
  1. R1Answers your business phone line 24/7. No business phone number or inbound call answering in the Grok app; voice agents are developer-API only.
  2. R2Books appointments during the call. No phone calls, so no booking during a call.
  3. R3Transfers calls to your cell when needed. No phone line, so no call transfer.
  4. E1Sorts your inbox and filters junk on its own. Email-triggered Automations and Gmail/Outlook connectors can label and move mail, but you write the triage rules yourself. source
  5. E2Drafts replies in your voice, waiting in your inbox. Connectors can create drafts and automations can respond to new mail; no ready-made 'drafts in your style', you set it up. source
  6. E3Joins your meetings and takes notes. No feature joins Zoom, Meet or Teams calls; it can summarize recordings you upload.
  7. E4Sends you task reminders. Scheduled Automations send you reminders by email or app notification. source
  8. E5Works with both Gmail and Outlook. Built-in connectors for Gmail and Outlook. source
  9. E6Replies to routine emails without you clicking send. An automation can respond when matching email arrives, but you build it yourself and send permissions must be enabled. source
  10. L1Finds leads from a built-in database. No built-in leads database; it researches prospects on the web and X. source
  11. L2Sends cold emails and follow-ups automatically. Can send email on a schedule via connectors and Automations; no built-in cold sequences, you build them. source
  12. L3Does LinkedIn outreach automatically. Grok Bot drafts LinkedIn outreach and its cloud browser can act on LinkedIn; no LinkedIn integration. source
  13. P1Writes SEO blog articles. Writes long-form blog articles in chat. source
  14. P2Publishes them to your website on its own. No CMS connector; a Grok Bot browser routine you build could log in and publish. source
  15. P3Keyword research and SEO audit. Can browse your site and research keywords on the web; no keyword-volume data or SEO crawler. source
  16. S1Writes posts in your tone. Writes social posts; Projects and custom instructions keep your tone. source
  17. S2Publishes to Instagram, TikTok, YouTube, LinkedIn, Threads, X and Facebook on a schedule. No social publishing connectors; only possible with a Grok Bot browser routine you build. source
  18. S3Creates images, carousels and videos. Grok Imagine generates images, multi-image sets and short videos. source
  19. W1Builds and hosts your website from a prompt. Grok Build creates a website from a prompt and hosts it, with custom domains. source
  20. W2Runs A/B tests on your site around the clock. No A/B testing feature for Grok Build sites. source
  21. W3Edits your website when you ask in chat. Grok Build sites are edited by chatting. source
  22. Q1Drafts quotes, estimates and invoices. Built-in skills create Word documents, spreadsheets and PDFs. source
  23. G1Reviews contracts and flags risky clauses. Upload a contract and it reviews it and flags risky clauses. source
  24. T1AI employees hand off work to each other. In Grok Bot (linked with SuperGrok), bots message each other and hand off tasks. source
  25. T2Group chat with your AI and human team. Grok Bot group chats are you plus bots; human teammates need Slack and a separate Cursor Teams seat. source
  26. T3Separate workspaces for each business. Projects give each client its own instructions and files. source
  27. T4Lets you build your own AI agent for any job. In Grok Bot, 'Create your own' makes a bot with its own job, apps and scheduled routines. source
Sintra · Sintra X · $48.50/mo
  1. R1Answers your business phone line 24/7. No phone number or voice agent; Milli writes call scripts only. source
  2. R2Books appointments during the call. No phone calls, so no booking during a call. source
  3. R3Transfers calls to your cell when needed. No phone call handling, so no transfers. source
  4. E1Sorts your inbox and filters junk on its own. Cassie's Inbox Manager labels every new Gmail/Outlook email into fixed categories, continuously. source
  5. E2Drafts replies in your voice, waiting in your inbox. Inbox Manager drafts replies for you to review, using your brand voice settings. source
  6. E3Joins your meetings and takes notes. Meeting Notetaker joins Meet, Zoom and Teams, but is an early beta for a limited number of users. source
  7. E4Sends you task reminders. No dedicated reminders; you can set up recurring scheduled tasks that email you. source
  8. E5Works with both Gmail and Outlook. Supports both Gmail and Outlook. source
  9. E6Replies to routine emails without you clicking send. Inbox Manager drafts replies, but auto-replying isn't supported; you approve each one. source
  10. L1Finds leads from a built-in database. No built-in leads database; Milli writes sales copy and strategy. source
  11. L2Sends cold emails and follow-ups automatically. Drafts cold emails and can send them via scheduled tasks; no built-in sequences, you set them up. source
  12. L3Does LinkedIn outreach automatically. LinkedIn integration manages a page and shares posts; no outreach. source
  13. P1Writes SEO blog articles. Seomi and Penn write SEO blog posts, including on a weekly schedule. source
  14. P2Publishes them to your website on its own. Pushes posts to WordPress as drafts for you to publish. source
  15. P3Keyword research and SEO audit. Seomi does keyword research and a site audit. source
  16. S1Writes posts in your tone. Soshie writes posts using your Brand Kit voice. source
  17. S2Publishes to Instagram, TikTok, YouTube, LinkedIn, Threads, X and Facebook on a schedule. Auto-publishes approved posts to Instagram, Facebook and LinkedIn; X is not supported and every post needs manual approval. source
  18. S3Creates images, carousels and videos. Generates images and carousels; no AI video generation. source
  19. W1Builds and hosts your website from a prompt. Commet builds and hosts a site from chat, with custom domains. source
  20. W2Runs A/B tests on your site around the clock. Help center says A/B testing is not supported for generated websites. source
  21. W3Edits your website when you ask in chat. Commet edits the hosted site when you ask in chat. source
  22. Q1Drafts quotes, estimates and invoices. The QuickBooks integration creates and sends invoices. source
  23. G1Reviews contracts and flags risky clauses. No dedicated feature, but you can upload a contract and ask a helper to flag risky clauses. source
  24. T1AI employees hand off work to each other. Helpers can delegate tasks to each other within a chat. source
  25. T2Group chat with your AI and human team. Team members share Brain AI, but helper chats are private to each person. source
  26. T3Separate workspaces for each business. Unlimited workspaces, one per business. source
  27. T4Lets you build your own AI agent for any job. The Agent Builder lets you create custom helpers connected to your apps, with scheduled tasks. source
ChatGPT · Plus · $20/mo
  1. R1Answers your business phone line 24/7. No business phone number or inbound call answering; 1-800-CHATGPT is OpenAI's own line for talking to ChatGPT.
  2. R2Books appointments during the call. No customer phone calls, so no booking during a call.
  3. R3Transfers calls to your cell when needed. No phone system, so no call transfers.
  4. E1Sorts your inbox and filters junk on its own. Can triage your Gmail or Outlook inbox on a schedule, but labeling and archiving wait for your approval. source
  5. E2Drafts replies in your voice, waiting in your inbox. OpenAI's inbox workflow learns your voice from sent mail and saves draft replies in Gmail or Outlook on a schedule. source
  6. E3Joins your meetings and takes notes. Record mode takes meeting notes on the Mac app, but doesn't join Zoom, Meet or Teams calls. source
  7. E4Sends you task reminders. Scheduled tasks send one-time or recurring reminders. source
  8. E5Works with both Gmail and Outlook. First-party Gmail and Outlook plugins. source
  9. E6Replies to routine emails without you clicking send. Can run a task when new Gmail arrives, but you set up the automation yourself and sending email can require your approval. source
  10. L1Finds leads from a built-in database. No built-in contacts database; web search and Agent can build prospect lists from public sources. source
  11. L2Sends cold emails and follow-ups automatically. Writes and sends emails from Gmail or Outlook when you approve each one; no automatic sequences or follow-ups. source
  12. L3Does LinkedIn outreach automatically. No LinkedIn integration; the built-in browser can act on signed-in sites when you confirm. source
  13. P1Writes SEO blog articles. Writes long-form, SEO-structured blog articles. source
  14. P2Publishes them to your website on its own. Webflow's app in ChatGPT can publish blog posts when you ask; one platform, not automatic. source
  15. P3Keyword research and SEO audit. Can audit a site by browsing it or through the Webflow app; no search-volume or ranking data. source
  16. S1Writes posts in your tone. Writes social posts in your tone from instructions and examples. source
  17. S2Publishes to Instagram, TikTok, YouTube, LinkedIn, Threads, X and Facebook on a schedule. Can't schedule posts itself; the browser can post on signed-in sites when you confirm. source
  18. S3Creates images, carousels and videos. Generates images and carousel slides; no video generation since Sora was shut down. source
  19. W1Builds and hosts your website from a prompt. ChatGPT Sites builds and hosts a site from a prompt, but it's in public beta and unavailable in the UK and EU. source
  20. W2Runs A/B tests on your site around the clock. Sites analytics shows visitors and page views; no A/B testing. source
  21. W3Edits your website when you ask in chat. Edits Sites websites in chat, and Webflow sites through Webflow's app. source
  22. Q1Drafts quotes, estimates and invoices. ChatGPT Work creates finished documents and spreadsheets you can send. source
  23. G1Reviews contracts and flags risky clauses. Upload a contract and it flags risky or non-standard clauses. source
  24. T1AI employees hand off work to each other. Splits a job across sub-agents of itself; named agents that hand off to each other are Business/Enterprise only. source
  25. T2Group chat with your AI and human team. Group chats were retired in July 2026; ChatGPT in Slack/Teams is Business/Enterprise only. source
  26. T3Separate workspaces for each business. Projects give each business its own chats, files, instructions and memory. source
  27. T4Lets you build your own AI agent for any job. Create your own plugins that combine instructions with connected apps, and run them on a schedule. source
Claude · Pro · $20/mo
  1. R1Answers your business phone line 24/7. No phone number or inbound call answering; voice mode only works inside the Claude app.
  2. R2Books appointments during the call. No phone channel, so no booking during a call (it can create calendar events in chat).
  3. R3Transfers calls to your cell when needed. No telephony features.
  4. E1Sorts your inbox and filters junk on its own. Gmail connector can label and organize threads, and a scheduled task can run it; no built-in always-on sorter, you set it up yourself. source
  5. E2Drafts replies in your voice, waiting in your inbox. Can save Gmail drafts and run a daily scheduled task to draft replies; you build the workflow and teach it your style. source
  6. E3Joins your meetings and takes notes. No meeting bot that joins Zoom, Meet or Teams; it can only read transcripts made by other tools.
  7. E4Sends you task reminders. Recurring scheduled tasks and calendar events can act as reminders; no native reminder feature. source
  8. E5Works with both Gmail and Outlook. Built-in Gmail and Microsoft 365 connectors (Outlook needs a work Microsoft 365 account). source
  9. E6Replies to routine emails without you clicking send. Can send Gmail replies, but asks before each send by default and you set up the scheduled task yourself. source
  10. L1Finds leads from a built-in database. No built-in contact database; it researches prospects via web search, or via paid third-party data connectors. source
  11. L2Sends cold emails and follow-ups automatically. Gmail connector can send and reply, and scheduled tasks can run a cadence you build; asks approval before sending by default. source
  12. L3Does LinkedIn outreach automatically. Claude in Chrome (beta) can be set up to do browser tasks on LinkedIn; no LinkedIn outreach feature. source
  13. P1Writes SEO blog articles. Writes SEO blog posts; the Marketing plugin has a blog drafting command. source
  14. P2Publishes them to your website on its own. Can publish through specific CMS connectors (e.g. Webflow, WordPress) plus a scheduled task you set up. source
  15. P3Keyword research and SEO audit. Marketing plugin runs an SEO audit using web search; real keyword volume data needs a separately paid connector. source
  16. S1Writes posts in your tone. Drafts social posts; the Marketing plugin includes brand-voice guidance. source
  17. S2Publishes to Instagram, TikTok, YouTube, LinkedIn, Threads, X and Facebook on a schedule. No native social publishing; only possible via Claude in Chrome (beta) browser tasks you set up. source
  18. S3Creates images, carousels and videos. Claude Design makes carousels and social layouts, but it can't generate photos or videos. source
  19. W1Builds and hosts your website from a prompt. Builds a site as an artifact and publishes it to a claude.ai link; no custom domain or real hosting. source
  20. W2Runs A/B tests on your site around the clock. No website A/B testing feature.
  21. W3Edits your website when you ask in chat. Edits artifact sites in chat, and Webflow sites via Webflow's official connector. source
  22. Q1Drafts quotes, estimates and invoices. Creates downloadable Word, Excel and PDF files. source
  23. G1Reviews contracts and flags risky clauses. The Legal plugin reviews contracts clause by clause with risk flags. source
  24. T1AI employees hand off work to each other. Cowork splits work across sub-agents of itself; not separate specialists handing off to each other. source
  25. T2Group chat with your AI and human team. No group chat with humans on Pro; Claude in Slack and shared projects are Team/Enterprise only. source
  26. T3Separate workspaces for each business. Projects act as separate workspaces with their own files, instructions and memory.
  27. T4Lets you build your own AI agent for any job. Create your own plugin that bundles instructions with connected tools, and run it on a schedule. source

Cite this report

Marblism Research (2026). AI Employee Index v1.0: Can AI tools run the back office of a service business? https://www.marblism.com/benchmark

Changelog

v1.0
October 7, 2026. First release: 6 products, 27 jobs.

In the 1 second it took you to get here, Marblism could have saved you an hour.

Meet Your AI Team