gofri Why two g's? When I was a kid my grandma called me Gofri, and the family picked it up. She said the G hard, almost twice, so I write it twice.
production

Multi-agent ARG platform

for
Anonymous agency, via ggofri
when
2024–25
Problem
An alternate-reality game that needed agents behind it.
My contribution
Built four human-supervised agent systems for an alternate-reality game; the agency's CEO later hired my next company for a follow-on build.
Decision
Keep people in control: reviewers can stop publication, restrict subjects, correct pace or context, or disable any agent.
Delivered
Four systems, built personally: journalist/newsletter, clue generation, investigator workflow and simulated players. Players were simulated.
Result
Four agent systems behind the game; the agency's CEO later hired my next company for a follow-on build.
Status
production · 2024–25

2 weeksper later system, reusing the first (about two sprints)

what I foundAgents publishing into a live game needed people who could stop, steer or switch them off.

group chat · 4 agents

  1. journalist Draft: tonight's newsletter on the missing archive. held by a reviewer
  2. clue writer Clue 7 ready: a date hidden in the photo caption. → players
  3. simulated player Has anyone else noticed the caption date? → players
  4. investigator Cross-checking the date against the archive log. → players
  • stop publication
  • restrict subjects
  • correct pace or context
  • disable an agent
Agents post, a reviewer decides what reaches playersreconstruction, illustrative

four systems

One platform, four agent systems

  1. 1Journalist / newsletterwrites the game's in-fiction reporting
  2. 2Clue generationproduces the clues players follow
  3. 3Investigator workflowagents that work the case alongside players
  4. 4Simulated playersappear as independent humans

control

People could stop it

  • Stop publicationa reviewer can halt what goes out
  • Restrict subjectstopics can be taken off the table
  • Correct pace or contextreviewers steer timing and framing
  • Disable an agentany agent can be switched off

reuse

The second system cost a fraction

  1. First systemabout two sprints, including the shared message bus and group chat
  2. Later systemsabout two weeks each, reusing it

disclosure

Simulated players, disclosed at the end

Simulated players appeared as independent humans; the fiction was disclosed at the end of the game. No engagement lift was measured.

ownership

Who did what, and when

  1. first systemabout two sprints
  2. later systemsabout two weeks each
  3. follow-on buildthe agency's CEO engaged my next company
  • MeAll four systems, the message bus and the reviewer controls
  • The agency's content reviewersSupervised the agents through the controls
  • FreelancersWorked on other startup engagements, not this one
  • Engagement measurementNone; I had no access to player metrics

What shipped

The agency's CEO later hired my next company for a follow-on build.

as reported; no engagement lift was measured
  • The agency reported satisfaction with the work, and its CEO later engaged my next company for a follow-on build.
Method and caveats

Interview testimony; I had no access to the agency's player metrics. Simulated players appeared as independent humans, with the fiction disclosed at the end of the game.

  • Scope and architecture: my own account, from interview testimony
  • The follow-on build: reported by the agency; no engagement lift was measured
  • The group-chat messages above are a reconstruction