Every strip below is a real build for a real client — the panels are the steps its graph actually runs, in order. Where a measured bill exists, the price on the frame is the measured one, not an estimate.
Reads each new applicant against the job's requirements, scores and summarizes them into HubSpot, and opens the SMS conversation with the candidate — with or without a resume attached. — 1,028 tasks (1,025 completed, 3 failed) in the 30 days to 2026-08-10 (04-agents/_probe/task_ranking_2026-08-10.json)
PRD_CV Screening Agent4.00 credits = $0.40 per candidate
1New-applicant webhook fires; extract the job's requirements2Branch on whether a resume came with the application3Extract and validate candidate details from the CV4Compare and score the candidate against the job; write the summary5Update the HubSpot contact with summary and job add-on6Send SMS 1 & 2 to the candidate (separate no-resume variant)7Generate the chat-thread ID (code) and create the SMS thread8Trigger the downstream scoring workflow
4.00 credits = $0.40 per candidate, exact menu match ('Screen a candidate' $0.375 + intake $0.025; Notion 'Favorite Staffing - Agent Cost Summary', 2026-02-27)(credit-rail measured)Screen a candidateEngage a candidateHubSpotSMS via custom APIcustom thread/scoring workflow APIs
04-agents/favorite-staffing/prd-cv-screening-agent/agent.json · Open the case file →
Parses every candidate SMS reply — confirmations, available hours, start dates, corrections — writes it back to HubSpot, keeps the conversation going, and hands qualified candidates to the qualification agent. — 1,154 tasks (1,150 completed, 4 failed) in the 30 days to 2026-08-10; 675 tasks in the 14 days to 2026-07-13
PRD_Reply Process AgentNo separately published per-run figure; the whole measured Favorite journey is $1.16 per candidate
1Inbound SMS reply arrives; branch on with/without-resume path2AI parses the reply: confirmations, hours, start date, corrections3Update the HubSpot contact with what the reply said4Send the last-availability SMS (SMS 3) or an additional chase for no-resume candidates5Trigger the qualification agent when the reply completes the picture6Trigger the scoring workflow on every exit path
No separately published per-run figure; the whole measured Favorite journey is $1.16 per candidate (11.59 credits across 4 agents, 6 executions), of which $0.15 (13%) is re-entry triggers between the split agents(credit-rail measured)Engage a candidateHubSpotSMS via custom APIcustom qualification/scoring workflow triggers
04-agents/favorite-staffing/prd-reply-process-agent/agent.json · Open the case file →
PRD_ Qualification Agent Favorite · 8 nodesIn production
Checks everything the candidate said over SMS against the resume as ground truth, issues the final qualification verdict, and writes it into HubSpot. — 506 tasks (505 completed, 1 failed) in the 30 days to 2026-08-10; 290 tasks in the 14 days to 2026-07-13
PRD_ Qualification Agent Favorite2.50 credits = $0.25 per candidate
1Validate the candidate's SMS replies against resume and summary — resume is the source of truth on conflict2Compress the running summary to under 2,000 characters3Update the HubSpot contact with the qualification verdict4Trigger the downstream scoring workflow
2.50 credits = $0.25 per candidate, exact menu match ('Qualify from the reply' $0.225 + intake $0.025)(credit-rail measured)Qualify from the replyHubSpotcustom scoring workflow API
04-agents/favorite-staffing/prd-qualification-agent-favorite/agent.json · Open the case file →
Scores every incoming candidate against the job's configured evaluation rules and posts the result straight into Zoho Recruit, announcing outcomes on Google Chat. — 3,919 screening tasks, 0 failed, in the 30 days to 2026-08-10 — second-highest volume on the platform (public case study: 9,500 candidates, 97.8% completion)
PRD_CVScreeningAgent~$0.05 per task at-cost
1Look up the job description and evaluation rules in the Baserow config table2Email the recruiter if the job has no rules row yet3AI scores the candidate against the configured criteria4AI determines match / no match5Code formats the result and posts it to Zoho Recruit6Announce the outcome on Google Chat
~$0.05 per task at-cost, billed — checked against the actual invoice within 1.3% (the 'simple screening agent' anchor in the finance deck appendix)(at-cost billed)Screen a candidateZoho RecruitBaserowGmailGoogle Chat
04-agents/booth-s-workspace-production/prd-cvscreeningagent/agent.json · Open the case file →
When a recruiter creates a new job in Zoho, turns the prose job description plus calibration notes into a structured must-have / nice-to-have rules row that the screening agent then reads — no engineer in the loop. — 146 tasks (140 completed, 4 failed) in the 30 days to 2026-08-10
PRD_EvalgenAgentno measured cost
1Zoho Recruit webhook fires on job creation; code extracts the job fields2Fetch existing rows from the Baserow config table; code checks recency3AI converts the prose JD + calibration notes into must-have / nice-to-have criteria4Code builds the config row payload5Create or update the row in Baserow for the screening agent to read
No measured cost yet — not billed through either railJob description to criteriaZoho Recruit (webhook)Baserow
04-agents/booth-s-workspace-production/prd-evalgenagent/agent.json · Open the case file →
PRD_TAInterviewAgent (instances 1 and 2) · 13 nodesIn production
Drafts and sends the interview invitation to a matched candidate using the recruiter's calendar details, from the correct regional careers mailbox. — 107 tasks (107 completed) on instance 1 plus 65 tasks (62 completed) on instance 2 in the 30 days to 2026-08-10
PRD_TAInterviewAgent (instances 1 and 2)no measured cost
1Pull the recruiter's calendar details from the Baserow interface table2Email back for missing information when the row is incomplete3Construct the job details and URL (code)4AI drafts the interview invitation email5Code assembles subject line and addressing6Send from the right regional mailbox (careers.co@ / careers.ph@hirebooth.com)
No measured cost yet — not billed through either railBook the interviewBaserowGmailZoho Recruit (job URL)
04-agents/booth-s-workspace-production/prd-tainterviewagent-1/agent.json · Open the case file →
PRD_Coolback-Order-Processing-Final · 79 nodesIn production
Turns inbound B2B order emails into validated, human-checked, ERP-ready orders — matching customers and delivery addresses against Airtable and filing every outcome back into the shared mailbox as a category tag. — 5,388 tasks (5,371 completed, 9 failed) in the 30 days to 2026-08-10 — the highest-volume agent on the platform; 2,721 tasks in the 14 days to 2026-07-13
1Classify email intent (order vs not) and tag the Outlook category2Validate the customer type; extract order-head details3Fetch customer records from Airtable; match the delivery address; tag exceptions ('PLZ nicht gefunden', 'Adresse nicht gefunden')4Attach the customer ID; extract the order line items5Merge everything and pause for the human-in-the-loop check6Generate the order CSV and upload via SFTP for ERP import7Archive the order PDF to DocuWare; append the log row to Google Sheets8Tag the mailbox 'Bestellung verarbeitet'
No measured cost yet — not billed through either railCase creationMicrosoft OutlookAirtableSFTP (ERP import)DocuWareGoogle SheetsS3
04-agents/coolback/coolback-order-processing-final/agent.json · Open the case file →
Works the shared fax/email order pool end to end: validates article codes and customers against the WinLine ERP, simulates the order, creates it after a human check, and files the original email to SharePoint — routing every exception to a named Outlook category. — 1,330 tasks (1,298 completed, 32 failed) in the 30 days to 2026-08-10; its companion PRD_Fraisa-Analytics-Agent logged 1,297 tasks (5 nodes, appends every outcome to Google Sheets)
PRD_Agent-AOP-Faxpoolno measured cost
1Intake from the fax/email pool; publish attachment URLs; set run variables2Categorize the email; route special cases to named colleagues' Outlook categories3Extract article codes; check prefixes in code4Extract buyer and order info; look up the customer in the WinLine ERP (custom API)5Validate each product against the ERP6Simulate the order; branch on errors, discontinued items, and orders over 10k EUR7Human check, then create the order in WinLine (CreateOrder V6)8File the order email to SharePoint as .eml; log every path to the analytics agent
No measured cost yet — not billed through either railCase creationProduct identificationAvailability and quantityMicrosoft OutlookWinLine ERP via custom APISharePointS3Google Sheets (via analytics agent)
TST_Wunsche_Aftersales_Agent · 50 nodesIn production
Handles consumer aftersales email — repair, replacement, spare parts, user manuals, service-center delegation — by classifying with confidence, routing on warranty status and claim value, and either replying from code-held templates or preparing a full handoff for human review. — 10 tasks (10 completed) in the 30 days to 2026-08-10; 20 tasks in the 14 days to 2026-07-13 — live but low-volume; the docs' clearest named production CX case
TST_Wunsche_Aftersales_Agent13 credits = $1.30 for a full logged run
1Look up case history in Airtable; parse the email, detect language, screen for escalation keywords2Identify sender type and retailer; query the product database (DAB)3Classify into a canonical category with confidence; extract product and customer facts4Route: repair/replacement by warranty status and claim value; spare parts; user-manual availability; non-service mail5Look up service-center and email-target contacts in Airtable6Assemble the reply — templates and signature held in code — or prepare the human-readable manual-review handoff7Create the case record in Airtable; send or reply; tag the mailbox
13 credits = $1.30 for a full logged run (~3 credits on early exit), measured August 2026 on a mid-tier model — documented as the 'CX aftersales full run' anchor; standard-tier menu equivalent is $0.88, so the gap is model tier(credit-rail measured)AftersalesTechnical documentsMicrosoft OutlookAirtableproduct database (DAB) via custom queryS3
04-agents/w-nsche/tst-wunsche-aftersales-agent/agent.json · Open the case file →
Turns an account manager's email into a correctly-formed Salesforce case: extracts the fields, verifies the contact and partner account, asks back only for what is missing, creates the case with attachments, and confirms with the case link. — 3 tasks (3 completed) in the 30 days to 2026-08-10; 0 tasks in the 14 days to 2026-07-13 — production wiring live but near-idle
PRD_OpsCaseHandlerno measured cost
1Intake the ADM's email; pull stored thread context from Airtable2Extract case fields; look up the contact and related accounts in Salesforce3Check completeness of required case fields; reply asking only for what is missing4Determine the relevant partner account5Create the case in Salesforce and attach files6Generate the case URL, confirm to the ADM by email, and log every step to Airtable
No measured cost yet — not billed through either railCase creationMissing-information chasingSalesforceGmailAirtableS3
04-agents/unidays-workspace/ops-case-handler-live-prod/agent.json · Open the case file →
ANOVA collections suite (PRD_ANOVADispatcher + PRD_ANOVACollector + PRD_ANOVAMissingData + PRD_ANOVAFinalConsolidator) · 39 nodesIn production
Runs daily receivables collection off SharePoint Excel: sends the day's dunning batches per the collection calendar, reads every reply, extracts payment data, chases missing information, writes the collection state back into the workbook cells, and consolidates the day's outcome. — Daily 09:34 UTC trigger; in the 30-day window to 2026-08-10: dispatcher 31/31 completed, final consolidator 31/31, missing-data chaser 30/30, collector 25/25 — 0 failures across the suite (a 2026-08-28 correction: it runs daily, not the earlier-reported 24 tasks per fortnight)
1Dispatcher (19 nodes) reads file index, collection calendar, SPOC list, state, and templates from SharePoint Excel; normalizes the calendar in code2Determines today's action per department against the dunning calendar and bulk-sends the reminder batches via Outlook3Collector (39 nodes, the main graph) catches replies; reads and normalizes SPOC master and collection state in code4Identifies the department, parses the attachment (Excel or not), and extracts the payment data5Writes updates back into the collection-sheet cells6Forwards to L2 or replies asking for exactly the missing data7Missing-data chaser (16 nodes) sends wave-based follow-ups; final consolidator (9 nodes) closes the day
No measured cost yet — not billed through either railDunning ladderMissing-document chasingSharePoint Excel (direct cell writes)Microsoft OutlookS3
Works the accounts-receivable follow-up queue: finds the next due task on each open case, drafts the dunning email, sends it through a Zendesk ticket set to the right brand and group, and announces the ticket on Slack. — 136 tasks (126 completed, 9 failed) in the 30 days to 2026-08-10
PRD_Betterment_AR_Agentno measured cost
1Find the open AR case and its next due task in the case tracker (custom API)2Code builds the inputs for the email3AI drafts the dunning email4Create the Zendesk ticket, set brand and group, and solve it to send the email5Mark the task complete and set the case to working6Announce the new ticket on Slack
No measured cost yet — not billed through either railDunning ladderZendesk via custom APISlackcustom case/task tracker API
Reads inbound insolvency correspondence for a receivables-management firm, identifies the sender, classifies each letter into its legal type (court letters, administrator letters, discharge, mandate changes, zero-plan), and sets the routing so the right desk gets it. — 237 tasks (237 completed, 0 failed) in the 30 days to 2026-08-10
PRD_INSO_Agent - 1~$0.23 per task at-cost
1Format the inbound correspondence (code)2Receptionist step identifies sender and contacts; code validates the contacts array3Primary classification into insolvency letter families4Specialist classification branches: administrator letters, court letters, discharge, mandate/dormancy, SBP and zero-plan, misc5Code sets the routing variables per class6Hand the classified case onward (FOA / BID system path)
~$0.23 per task at-cost, billed on a private instance — the 'complex regulated agent' anchor in the finance deck appendix, named as BID in the rate canon(at-cost billed)Case creationCollections follow-upEmail intake (Beam trigger)prompt classifiers + Code Executor only — no external system writes in this graph
Validates a travel expense claim against company policy through six named validators (cabin class, hotel and rental caps, receipt financials, calculations and tax, international routing), then returns a rendered PDF validation report to the submitter by email. — 1 task in the 30 days to 2026-08-10 — no steady volume
PRD_TravelExpenseAgentno measured cost
1Email intake; extract structured trip data and receipt data2Run six named prompt validators: policy V-001+B-003, hotel/rental B-004+B-005, receipt financials, calculations & tax, international routing3Generate the validation report and evaluation matrix4Render the HTML report in Python and convert to PDF5Upload to S3 and reply to the original email with the PDF attached
No measured cost yet — not billed through either railPolicy complianceDocument verificationGmailOneCompiler (Python)PDF renderer via custom APIS3
Backfills campaign pacing trackers from ad-platform exports: code computes all the spend and impression figures, AI touches only new ad-set names, and the write-back changes exactly three manual cells per row — never a formula cell. — 18 sandbox tasks (14 completed, 4 failed) in the 30 days to 2026-08-10 — test runs, not client volume; the media line docs state none of the five media scenarios has daily production volume yet
SBX_Adenva Social Pacingno measured cost
1Parse platform export CSVs and tracker state; flag stale dates before anything runs (code)2Deterministically derive YTD, yesterday, and last-7-days spend and impressions per ad set (code)3Match ad-set names against the persistent mapping table; only new names go to AI (code)4AI maps new or drifted names to template rows with confidence and evidence5Code writes only the 3 manual cells per row by row type, skipping actualized and formula rows6Send an audit-style run note and completion email reporting every exception with its reason
No measured cost yet — not billed through either railCollect filesNormalise formatResolve namesCheck pacingWrite back and draftAd-platform CSV exportsExcel pacing trackers (17 workbooks at the real client)email
Answers inbound calls from placed candidates, looks up their record without ever reading its contents aloud, and drafts every follow-up action for a human to approve. — 203 calls in the measured cost sample; 179 minutes across 51 calls (3.51 min/call) in the duration sample — demo traffic, not client production; tools run via browser + Cloudflare Worker, not yet connected to a Beam agent
Candidate Care Line (inbound voice demo)13.1 credits = $1.31 per run
1Answer the inbound call (outbound calling is explicitly excluded)2Look up the candidate's record; disambiguate with questions that reveal nothing from the file3Handle the request conversationally, never reading record contents aloud4Draft any resulting action as a proposal for human approval5Hand off a post-call summary
13.1 credits = $1.31 per run, measured across 203 calls on a mid-tier model (standard-tier menu price is $0.855 — the entire gap is model tier); voice metering $0.1115/min on top at 3.51 measured min/call(credit-rail measured)Post-placement careSynchronous voice (channel)Telephony (demo stack)Cloudflare Worker to spreadsheetsbrowser tools — not on the Beam agent platform
All production statuses and volumes come from the task probe (04-agents/_probe/task_ranking_2026-08-10.json, 30-day window 2026-07-11 to 2026-08-10; 197 agents scanned), cross-checked against the 14-day Notion snapshot of 2026-07-13 (03-projects/alloyed-cx/04-outputs/production-cx-table-zh.md) — never from the graph files' isPublished/isEverExecuted flags, which the workspace has proven unusable (coolback shows all-false flags against 5,388 probed completions). Node counts and step lists are read from the 2026-08-28 graph snapshots in 04-agents/, which prove graph shape only; re-verify against the live published graph before quoting to a client. Costs sit on two rails an order of magnitude apart: credit-rail menu prices (1 credit = $0.10 assuming 100k tokens/call) are ceilings, and at-cost enterprise billing lands 5-10x lower — quote the rail the contract actually uses. Two calibration points match the menu exactly (Favorite screening $0.40, qualification $0.25); every measured deviation (Wünsche $1.30 vs $0.88, Hershey $1.31 vs $0.855) is model tier, not node count. The unattributed '~$0.80/task very heavy multi-control agent' at-cost anchor is deliberately pinned to no card. BID Coburg straddles lines: filed under cx per the production-CX grouping, while its $0.23 at-cost figure is the finance deck's 'complex regulated agent' anchor. Cards with cost_basis 'none' (Coolback, Fraisa, Booth non-screening, Americana, Betterment, UNiDAYS, MHP, CMI pacing) have no per-agent measured figure anywhere in the repo; the honest fallback is the line deck's menu price plus the at-cost band. Media has no production volume on any of its five scenarios — its run-rate proof (Americana daily suite) is borrowed from another industry and proves workflow shape, not media experience.