Agents in the Wild

This directory is no longer updated. Last update: November 16, 2026.

Hostinger's DEX handles about 17,000 requests a month, and staff built 182 custom agents on it

Hostinger turned its internal Slack assistant, DEX, into a platform where any team can build a persistent agent. The company also says what still goes wrong: hallucinated answers and risky actions.

Organization
Hostinger
Industry
Web hosting
When
May 2026
Evidence
Company reported

Reported outcome

Around 600 weekly users, roughly 17,000 requests a month across 11 departments, and 182 custom agents as of May 2026. Weekly replies grew from around 3,500 to over 10,000.

  • around 600 (500 return three or more times)
    Weekly users
  • roughly 17,000 across 11 departments
    Requests per month
  • 182
    Custom agents (May 2026)
  • around 3,500 to over 10,000
    Weekly replies, February to May 2026
  • roughly 30
    Connected systems (early 2026)

Caveats

Stated by the company that uses the agent (blog, earnings call, interview). Not independently audited.

  • Hostinger wrote this about its own tool. Nobody outside checked the numbers.
  • These are usage numbers. The post gives no accuracy rate and no time saved across the company.
  • Hostinger says hallucination is still one of its biggest problems: DEX sometimes gives a confident answer when it should say it does not know.
  • One featured agent, the Jira ticket closer, was still in testing and had not touched real releases.
  • The post does not name the AI model.
  • Disclosure: this site is hosted on Hostinger. Hostinger did not pay for, ask for, or see this entry.

What the agent does

DEX started in late 2024 as a Slack button that asked engineers to rate their developer experience. An AI layer came in mid-2025. In autumn 2025 DEX started to plan before it acted and to pick the right tools for each request. By early 2026 it was connected to roughly 30 systems, including Jira, GitHub, Google Docs, calendars and incident management.

People still used it the same way: ask, get an answer, leave. So in early 2026 the team shipped custom agents. A person builds an agent once, with its workflow, format and defaults. Agents can also run on a schedule. One posts a weekly performance summary to its team's channel every Monday.

What happened

Engineers, growth, content, QA, product and design teams built 182 agents. One agent rotates the on-call owner and updates the Slack group tag. One growth report agent, by its owner's account, showed 40% of performance the team had not tracked before.

Hostinger also lists what it has not solved. Acting is riskier than answering, so the ticket-closing agent stays in testing. DEX cannot read private Slack channels. Rate limits are shared, so one heavy agent can slow the others.

What to copy

  • Start as a helper inside existing workflows, and add intelligence later.
  • Let people save a prompt as a named agent so the team stops retyping it.
  • Test any agent that changes records before it runs on real work.

Starter prompt

Written by us, not by Hostinger. Test on a copy before you let it close real tickets.

Release notes: {release}
Open tickets: {tickets}

Match each released change to its ticket. Show the evidence for each match (PR link, ticket key).
List tickets you would close, and tickets you are not sure about.
Do not close anything. Wait for a person to confirm the list.
If you do not have the information, say "I don't have that information."

Sources

  1. DEX case study: From internal chatbot to AI agent platform — Hostinger, May 28, 2026

Added Nov 16, 2026 · Sources last checked Nov 16, 2026.

Get new use cases weekly

One short email. Sources included.

The newsletter is paused. Past issues stay in the archive.

Related use cases