Most AI agency websites show concept slides. We show production agents, three of them, with names, full specs, eval pass rates, and integration details. These are the same agents we customize and deploy in client engagements. Names changed to protect client data; mechanics are real. For the full catalogue of 14 agent types we build across the three organisational levels, see AI across three organisational levels.
theagency47 maintains three production AI agents as reference deployments: Sofia (sales outreach, Operational tier), Yiannis (customer support, Operational tier), and Maria (email triage, Task tier). Each is built on Anthropic Claude, has a full one-page specification, a 20+ case eval suite with 90%+ pass rate, real tool integrations (Gmail, Slack, HubSpot, Zendesk), and is live in production. When clients sign Spark or Workforce engagements, we typically use one of these three as the starting template, customizing the system prompt, knowledge base, and integrations to the client's specific business. The agents are showcased here as proof-of-work, not as products you can buy off-the-shelf.
theagency47 · Updated May 2026What she does: Researches prospects, drafts personalized cold outreach in your voice, schedules follow-ups, hands hot replies to a human. Eats 8 hours of SDR work per day, surfaces the 30 minutes of human judgment that still matters.
What he does: Reads inbound support tickets, resolves tier-1 questions (order status, account access, basic troubleshooting) autonomously, escalates the rest with full context, maintains the FAQ knowledge base from new patterns he sees.
What she does: Watches your inbox in the background. Categorizes mail by urgency and topic, drafts replies to common patterns (intro requests, scheduling, vendor questions), surfaces only what actually needs you. Reclaims 60–90 minutes per day.
Sofia, Yiannis, and Maria are the three we show in full. They are not the only three we build. Here is the complete catalogue of AI agents and automations we deploy for clients, grouped by the organisational tier each one belongs to. Every one is built on Claude, wired into your tools, and shipped with an eval suite.
Resolves tier-1 tickets across chat, email, and WhatsApp from your knowledge base. Typically clears 50–70% of volume autonomously, escalating the rest with full context. Yiannis is the live showcase.
Researches prospects, writes personalised outreach in your voice, sequences follow-ups, and books meetings. Built for B2B teams with a defined ICP. Sofia is the live showcase.
Builds target lists from your ICP, enriches them with firmographic and contact data, and runs first-touch personalised outreach, from scraping to send.
Handles inbound and outbound calls in natural Greek or English (appointment confirmations, lead qualification, customer-feedback calls) with hand-off to a human when needed.
Answers your phone line around the clock, handles routine questions, captures details, and transfers live to a human with the full conversation as context.
Books, reschedules, and confirms appointments over chat, email, or call, synced to your calendar, with reminders and no double-bookings.
Enriches records, scores leads, logs activity, and drafts the next follow-up so your pipeline never goes stale between touches.
Takes a brief to a publish-ready draft (blog posts, social, newsletters) in your brand voice, with internal-link discipline and SEO structure.
Keyword research, content briefs, and content-gap analysis that feed your editorial calendar, the research layer behind a content engine.
Turns raw data from your tools into a clean executive summary on the schedule you set (daily, weekly, or month-end) ready to forward.
Answers ad-hoc questions over your own data in plain language. No SQL, no waiting on an analyst, ask, get the chart and the takeaway.
A role-specific assistant trained on one person's processes, documents, and decisions, a force-multiplier for a founder, partner, or department head.
Categorises your inbox by urgency and topic, drafts replies to common patterns, and surfaces only what needs you. Maria is the live showcase.
Drafts, sends, and follows up on transactional and outreach email from templates and triggers, the sending layer behind a workflow.
A site or WhatsApp chatbot grounded in your knowledge base, answering in Greek and English, with a clean handoff to a human when it hits its limits.
OCR, extraction, and classification of invoices, contracts, and forms into structured records your systems can use.
Builds and maintains a searchable knowledge base for your team or customers from the documents you already have, and keeps it current.
Connects your tools (n8n, Make, Zapier, or custom) so no one copies data between systems by hand again. Usually the fastest ROI of the set.
Workflows, a prompt library, and team training so your people use frontier models safely and effectively in their day-to-day.
Purpose-built tools for when off-the-shelf doesn't fit the workflow, scoped, built, and owned by you.
End-to-end delivery from proof-of-concept to production, evals, integrations, monitoring, and the handover docs to run it.
The three showcase agents are reference deployments that demonstrate the shape of work we do. When you sign an engagement, we use the relevant showcase as a starting template and customize it to your business. Specifically:
If you want to see what a custom version of Sofia/Yiannis/Maria would look like for your business, that is exactly what the discovery call is for, or you can describe the agent you have in mind in plain language and we'll send back a structured brief. For the broader catalogue of 14 named agent types we build across the three tiers, see AI across three organisational levels.
Each showcase agent is deliberately positioned at one of the three tiers of our workforce framework, so you experience all three when reviewing the work.
Decision-support agents that read your data and produce briefings. We build these case-by-case during engagements because they are too client-specific to template (your "executive brief" is not the same as anyone else's). Examples we have built privately: weekly competitive-intel digest, pipeline health analyst.
Departmental-flow agents. Sofia handles sales outreach, Yiannis handles tier-1 customer support. Both are productizable enough to template, customizable enough to be useful per business.
Maria handles email triage, the work no team actually wants to own. Task agents are the highest-volume, lowest-customization tier; the same agent template fits many businesses with minor tuning.