
- I run five automated "bots" across my companies. By the strict definition only one is an AI agent (Emily, GAPLI's phone assistant), one is an AI workflow (the article pipeline behind madrydochod.pl) and three use no AI at all: they are the counters and checkers that tell me whether the AI ones work.
- Emily, live since 9 Oct 2026: 305 calls by 27 Oct 2026; 63 of 136 inbound calls resolved (46%), but only 30 of 169 outbound calls became a conversation (18%) and 45 went to voicemail.
- The article pipeline: 6 new articles a day since 15 Oct 2026, with a written brake to 4 a day if more than 40% stay unindexed for 2 weeks; 141 posts were live by 26 Oct 2026.
- The non-AI guards: a /go/ click counter that counts only humans (a partner network recorded 153 clicks with about 1 real reader), a server-log traffic counter, and a checker that tests 12 banks' login pages every 5 minutes.
Contents · 8 sections
- 01AI agent examples: which of our five bots are really AI agents?
- 02AI agent example 1: Emily, an AI phone assistant for a dropshipping platform
- 03AI agent example 2: the article pipeline behind madrydochod.pl
- 04Automation examples 3 to 5: the scripts that keep the AI honest
- 05AI agent examples: what do the tools behind them cost?
- 06How do you pick your first AI agent example to copy?
- 07Frequently asked questions
- 08Sources
This is for you if you run a store or a small company and every list of "AI agent examples" is vendor demos. Below are five automations that run in my businesses, with their logs and an honest label: real agent, AI workflow, or plain script.
Emily answers the phone for GAPLI and is built in Telofia, both my companies, so read the disclosures below. I also show where each tool is not what I would recommend. The full call funnel, including the outbound calls that did not work, is in AI phone assistant: 305 calls, 46% resolved inbound.
AI agent examples: which of our five bots are really AI agents?
Of the five automations running in my companies on 3 Nov 2026, one is an AI agent (it decides its next step and uses tools), one is an AI workflow (a model fills fixed steps) and three are plain rule-based scripts with no language model. I use Anthropic's definition from "Building effective agents" (19 Dec 2024) to sort them.
Workflows follow predefined code paths; agents direct their own process and tool use. What is an AI agent? has the longer explanation.
| Example | Where | Type | Key number (date) | My verdict |
|---|---|---|---|---|
| Emily, AI phone assistant | GAPLI (phone, browser, Telegram) | AI agent | 46% of inbound calls resolved (27 Oct 2026) | Keep inbound, rethink outbound |
| Article pipeline | madrydochod.pl, TillRadar, MainStreetRadar | AI workflow | 6 articles a day (since 15 Oct 2026) | Works, watched by an indexing brake |
| /go/ human-only click counter | madrydochod.pl partner links | Rule-based script | 153 network clicks vs about 1 reader | Essential, cheap |
| Server-log traffic counter | All sites, the Empire Counter | Rule-based script | 80 people in 9 days on madrydochod.pl (17–25 Oct 2026) | My main source of truth |
| Bank availability checker | kontowtescie.pl | Rule-based script | 12 banks every 5 minutes (since 25 Oct 2026) | Too early to judge |
I list the three scripts because without them I could not give one honest number about the two AI systems.
AI agent example 1: Emily, an AI phone assistant for a dropshipping platform
Emily is GAPLI's AI assistant, built in Telofia and live since 9 Oct 2026 on the phone, in the browser and on Telegram; by 27 Oct 2026 she had handled 305 calls (13 test calls excluded), resolving 63 of 136 inbound calls (46%) while only 30 of 169 outbound calls became a conversation (18%).
She is an agent in the strict sense: during a call she chooses to answer, open a ticket, request a callback or hand over to a person (9 tickets and 1 handover inbound). Before launch she scored 85/100 in Telofia's pre-launch test, with a knowledge base of 189 FAQs and 239 passages from 49 pages.
Outbound is the weak side: 94 of 169 calls were not answered and 45 went to voicemail, and onboarding progress was 2 of 24 for operators who talked versus 2 of 28 for those who did not. The full call table is in What is an AI agent? and the cost math in How much does an AI agent cost?
If your customers mostly need written answers, a good help page costs nothing per minute. If your main goal is outbound calling, my 18% says the phone is a hard channel. If you have developers, you can wire public voice and carrier services yourself (prices below). If you want to try a ready assistant, Telofia is here, and the store platform Emily works for is GAPLI.
AI phone agents and the law: TCPA in the US, PECR in the UK
In the US, the FCC ruled on 8 Feb 2024 (FCC 24-17) that the TCPA's limits on "artificial or prerecorded voice" cover AI-generated voices, so such calls need the called party's prior express consent unless an emergency purpose or exemption applies. In the UK, regulation 19 of PECR bans automated marketing calls with recorded matter unless the subscriber has consented and the caller's number is shown.
Emily calls store operators in Poland, so these rules do not govern her. If you copy the pattern for US or UK numbers, get consent first, keep a "do not call" list and say it is an AI. This is not legal advice; read the texts in the sources or ask a lawyer.
AI agent example 2: the article pipeline behind madrydochod.pl
The article pipeline is an AI workflow, not a full agent: it publishes 6 new articles a day on madrydochod.pl since 15 Oct 2026, following fixed steps and a written brake rule that cuts the pace to 4 a day if more than 40% of posts stay unindexed for 2 weeks. By 26 Oct 2026 the site had 141 published posts in 20 days.
The steps are fixed: topic from a calendar, draft, automatic checks, scheduled publication. The model writes inside them and decides nothing else, which makes it easy to audit at this pace.
On 15 Oct 2026, 41% of madrydochod.pl posts were not indexed by Google (Search Console), which is right at the brake line. From 17 to 25 Oct 2026 the server logs counted 80 people in 9 days (3 to 18 a day), and 27 of them (34%) came from ChatGPT. The same pipeline was ported to the UK (TillRadar, 28 posts by 26 Oct 2026, 0–5 people a day) and the US (MainStreetRadar, 12 scheduled posts on 26 Oct 2026).
Volume is easy for a model; demand is not, and the brake rule matters more than writing speed. For AI-search citations, see How to get your site cited by ChatGPT.
Automation examples 3 to 5: the scripts that keep the AI honest
The other three examples contain no AI: a /go/ link counter that counts only human clicks, a server-log counter that counts only people, and a checker that tests 12 Polish banks' login pages every 5 minutes. They matter because on madrydochod.pl a partner network recorded 153 clicks when about 1 real reader had clicked.
The /go/ human-only click counter
Partner links on madrydochod.pl go through a /go/ address. A network's panel once showed 153 clicks while about 1 real reader had clicked; the rest were bots. That is why our own /go/ counter counts only humans, with plain rules and no model, and no fake engagement ends up in my reports.
The server-log traffic counter
My counter reads the server logs and separates people from bots; on kontowtescie.pl, for example, a visit counts as a person only when the browser also sent a small ping file from the page. On 8 Oct 2026, two days after launch, madrydochod.pl had zero visits from Google and social, while Googlebot fetched 53 pages, GPTBot 78 and ClaudeBot 181. The same method feeds the live Empire Counter for this site and the others:
| Site | People | Page views | From Google and Bing | From ChatGPT and other AI |
|---|---|---|---|---|
| The Online Empire gregciupek.com · this site, English, since 28 Oct 2026 | 149 | 167 | 45 | 24 |
| Internetowe Imperium (PL) internetoweimperium.pl · Polish edition of The Online Empire, since 26 Oct 2026 | 37 | 76 | 10 | 0 |
| Mądry Dochód (Poland) madrydochod.pl · affiliate blog, since 6 Oct 2026 | 112 | 153 | 33 | 37 |
| TillRadar (UK) tillradar.co.uk · UK business banking, since 20 Oct 2026 | 136 | 149 | 67 | 19 |
| Konto w Teście (Poland) kontowtescie.pl · bank reliability index, since 26 Oct 2026 | 12 | 16 | 1 | 4 |
| MainStreetRadar (US) mainstreetradar.com · US business banking, since 27 Oct 2026 | 66 | 69 | 25 | 7 |
| Dziennik Firm (Poland) dziennikfirm.pl · Polish business daily from the company register (KRS), since 5 Nov 2026 (counter since 6 Nov, counts only people who accepted statistics) | 138 | 326 | 118 | 3 |
| Companies Daily (UK) companiesdaily.co.uk · UK business daily from Companies House, since 6 Nov 2026 (counter since 7 Nov, counts only people who accepted statistics) | 22 | 39 | 4 | 10 |
| Companies Daily US companiesdaily.com · US business daily from SEC EDGAR filings, since 11 Nov 2026 (counter since 12 Nov, counts only people who accepted statistics) | 10 | 11 | 10 | 0 |
| Total | 682 | 1,006 | 313 | 104 |
Source: web01 server logs, counted by a script every 30 minutes. A person is a browser that ran our script; bots, data-centre addresses and our own visits are removed. Google Analytics shows less because it runs only after consent. How we count
The full breakdown by site is on the Empire Counter page.
The bank availability checker on kontowtescie.pl
kontowtescie.pl, a reliability index of Polish banks public since 25 Oct 2026, checks 12 banks' login pages every 5 minutes and adds app ratings and decisions of the consumer watchdog UOKiK. The first full index is due on 24 Nov 2026. If you only need to know when your own site is down, you do not need to build this: UptimeRobot's free plan, checked on 3 Nov 2026, includes 50 monitors at a 5-minute interval.
AI agent examples: what do the tools behind them cost?
On official price pages checked on 3 Nov 2026, a voice agent layer costs $0.08 a minute (ElevenLabs Speech Engine agents), US phone carriage $0.0085 a minute to receive and $0.0140 a minute to place a local call, plus $1.15 a month for a local number (Twilio). Simple uptime checks can be free.
| What you would need | Tool and item | Price |
|---|---|---|
| Voice for a phone agent like Emily | ElevenLabs, Speech Engine agents, per minute | $0.08 |
| Receiving calls | Twilio, local US number, per minute | $0.0085 |
| Placing calls | Twilio, call to a US local number, per minute | $0.0140 |
| Phone number | Twilio, local US number, per month | $1.15 |
| Availability checks like kontowtescie.pl | UptimeRobot Free: 50 monitors, 5-minute interval | $0 |
| Human-only click and traffic counters | Own code on your server | your time |
ElevenLabs also lists "Realtime" speech-to-speech models at $0.08 a minute plus model usage, so check what you buy. Token prices are compared in How much does an AI agent cost?; none of these are my invoices.
How do you pick your first AI agent example to copy?
Copy the measurement first and the agent second: in my fleet the two AI systems became judgeable only after three non-AI counters existed. Then start with one inbound, repeatable task, as with Emily's 46% resolved inbound calls, rather than outbound calling, where she reached a conversation in 18% of 169 calls.
- Build the counter before the agent. Count real people, clicks and results, without bots and without yourself.
- Pick one inbound job. Questions you already answer by hand every week.
- Write the brake rule in advance. Mine: 4 a day if more than 40% stay unindexed for 2 weeks.
- Check the law before outbound. TCPA in the US, PECR in the UK.
No sales are attributed to Emily's calls by 27 Oct 2026, and the sites fed by the article pipeline recorded their first conversion only on 23 Oct 2026: one business-account application with our partner code on madrydochod.pl. More tools and tests are in the AI & automation topic.
Frequently asked questions
What is a real example of an AI agent in a small business?
Emily, the AI phone assistant of GAPLI, is one. Between 9 and 27 Oct 2026 she handled 305 calls, resolved 63 of 136 inbound calls (46%) and opened tickets or handed over to a human when she could not help. She decides those steps herself during the call, which is what makes her an agent.
Is an automated content pipeline an AI agent?
Usually not. If the steps are fixed and the model only writes inside them, it is an AI workflow by Anthropic's definition, like my madrydochod.pl pipeline at 6 articles a day since 15 Oct 2026.
Do AI phone agents need consent in the US and UK?
For many calls, yes. The FCC confirmed on 8 Feb 2024 that AI voices fall under the TCPA's artificial-voice consent rules, and UK PECR regulation 19 requires consent for automated marketing calls.
Which AI agent example should a store owner copy first?
An inbound question-answering assistant, after you have a counter for real results. In my data inbound worked far better (46% resolved) than outbound (18% reached a conversation). If written answers are enough, a help page is cheaper than any agent.
Sources
- GAPLI database, Emily call records 9–27 Oct 2026, 13 owner test calls excluded (read 27 Oct 2026); Telofia pre-launch test report (9 Oct 2026).
- madrydochod.pl, kontowtescie.pl, tillradar.co.uk and mainstreetradar.com: WP REST post counts (26 Oct 2026), server-log counters (8 Oct and 17–25 Oct 2026), Google Search Console (15 Oct 2026).
- Anthropic, Building effective agents (19 Dec 2024, checked 3 Nov 2026): anthropic.com
- Federal Communications Commission, Declaratory Ruling FCC 24-17 on AI technologies and the TCPA (released 8 Feb 2024, checked 3 Nov 2026): docs.fcc.gov
- UK legislation, The Privacy and Electronic Communications (EC Directive) Regulations 2003, regulation 19 (as amended 2016, checked 3 Nov 2026): legislation.gov.uk
- ElevenLabs, API pricing (checked 3 Nov 2026): elevenlabs.io
- Twilio, Programmable Voice pricing for the United States (checked 3 Nov 2026): twilio.com
- UptimeRobot, Plans and pricing (checked 3 Nov 2026): uptimerobot.com
As of 3 Nov 2026. Emily's figures are from the GAPLI database as read on 27 Oct 2026; site figures are from the dates given in each section. The ElevenLabs, Twilio and UptimeRobot price pages, the FCC ruling, PECR regulation 19 and the Anthropic article were checked on 3 Nov 2026. I will add the first full kontowtescie.pl index after 24 Nov 2026.



