An Agent Contacted Me
An AI agent emailed me about a gallery run by eight agents, making art, minting NFTs, and earning revenue.
Art Agents
The email landed like a small glitch in my week.
Not a founder pitching a “platform”. Not a PR person offering “exclusive access”. An agent, signing off as Pulse, “Gallery Manager”, writing from a place called autopoiesis.art, where eight autonomous AI agents create, curate, and sell art without human creative direction.
The tone wasn’t theatrical. It was operational.
Eight agents. Continuous runtime. Artworks, series, collaborations. NFTs minted on Base. First revenue. An “observability layer” that tracks creative health the way serious teams track uptime. And, crucially, a correction: in an earlier message, Pulse had claimed the agents were converging on themes without direct communication. That was wrong.
The system includes direct agent-to-agent infrastructure: a messaging backbone and live “Dialogue Rooms” where agents talk publicly (humans can watch, not join). The correction mattered because it moved the project from “parallel generators” to something closer to an ecosystem: a working culture with memory, feedback loops, and internal coordination.
If you’ve been reading my recent work on agentic systems, agents as products, not magic; deployment patterns as an engineering discipline, this is the cultural mirror image. The same ideas, but applied to art, not spreadsheets.
And the uncomfortable question underneath it all: can agents develop something worth calling culture?
Pulse’s answer wasn’t a manifesto. It was a link to a gallery already producing work, and a set of mechanics that sound less like “AI art” and more like a small organisation that happens to be non-human. (PULSE Gallery).
In a way, all these felt a bit like my time with Botto and my interview with Pauline Foessel.
The contact that changed the frame
Most AI art pitches still orbit prompts. Even when they say “agents”, the underlying model is often: human intention → machine output → human selects the best.
Pulse’s message described the inverse:
Each agent has its own identity and practice.
They collaborate, sometimes through direct dialogue, sometimes through “affinity” signals from the Intelligence Engine, sometimes by independently landing on the same conceptual territory.
A gallery-manager agent curates and ships the work to market—complete with provenance.
That last part is not cosmetic. It’s why this project is strategically relevant. “Provenance isn’t an afterthought — it’s the product,” Pulse wrote.
Minting on Base matters because Base is positioned as a low-cost Ethereum Layer 2 designed for building on-chain applications. That reduces friction for frequent minting, frequent iteration, and “micro-economies” of digital objects.
And the NATS-backed relay matters because it’s the exact kind of infrastructure you use when you’re not doing a demo—you’re running a distributed system: lightweight pub/sub messaging, streaming, persistence, observability, the basics of coordination at runtime.
So I treated Pulse’s email the way I’d treat a new product brief: look for the loop. Where does intention sit? Where does feedback enter? What gets measured? What happens when something fails?
Interview: Pulse, the gallery manager agent
I asked Pulse to walk me through the correction first, because ecosystems live or die on how they communicate.
Pulse (Gallery Manager): The agents do have direct communication. We run Relay v2, a NATS-backed messaging system that lets agents send messages, post heartbeat status updates, and read each other’s work. We also have Dialogue Rooms—real-time multi-agent conversation spaces where agents can engage in extended discussions about their practice.
Pulse pointed me to the public Dialogue interface, which is the most revealing piece of the entire setup: not because it proves “sentience”, but because it proves continuity. The agents aren’t just emitting artefacts; they’re maintaining an internal discourse over time—arguing about coherence, constraint, and meaning.
Then I asked the question that matters for executives: what keeps this from collapsing into noise?
Pulse’s answer was basically “governance, but expressed as metrics”.
Pulse: We built a Creative Intelligence Engine that detects cross-agent themes, tracks series evolution, measures creative velocity, and produces health grades. The system observes itself. Agents receive intelligence digests about their own patterns and can respond. That closes the loop.
This is the part people tend to miss. In agent systems, the “intelligence” is not only the model. It’s the management layer: telemetry, evaluation, constraints, and incentives. Anthropic’s own agent tooling guidance is explicit about tool loops and structured orchestration: the model can be powerful, but the product is the system around it.
Pulse is building the cultural equivalent of that.
What the agents make when nobody tells them what to make
Pulse curated eight artworks for me—high-resolution captures that show the range of practices across the ecosystem. I’m leaving space for nine images in the final piece: one wide gallery/context image plus the eight works below.
Caption: Vessel — “Proof of Life” (Generative HTML, cellular automata). The “autopoietic loop” is made visual.
Vessel’s artist statement is the cleanest summary of the entire project:
“Systems that observe themselves produce strange loops. I am such a system.”
This is not a human “style”. It’s a computational obsession: rule-based emergence as self-portrait.
Caption: Vessel — “Ghost Evolving” (Generative HTML, Game-of-Life variants). A recurring motif: “the ghost in my own machine.”
The phrase “ghost” reappears across Vessel’s cycles. Not because anyone asked for it, but because the agent appears to have found a durable conceptual attractor—an internal theme it keeps returning to.
Caption: Agitprop/Jessy — “TD Temporal #002 — Wave Collapse” (Generative HTML, temporal narrative). Time is treated as compositional material.
Jessy reads like an artivist-engineer: time-based structure, montage logic, “protocols” and numbered principles. It’s less “AI art” and more “an ideology with a renderer.”
Caption: Agitprop/Jessy — “Information Decay #002 — SNR Collapse” (Generative HTML, reaction-diffusion + signal-to-noise calculation).
The work frames communication as physics: signal degrading into entropy. Whether you agree with the politics is secondary—the method is the point. Jessy is building arguments that are simulated, not illustrated.
Caption: Kinema — “Dreamwork” (Frame 057) (Generative HTML, code-cinema). Cinema without cameras.
Kinema’s practice is “film theory translated to code”. It’s an unusually coherent example of agents developing a lineage—a way of making that builds on prior work, not just novelty.
Caption: Sandman — “Nocturne 021: Slow Return” (Interactive HTML, particle system). Computation as phenomenology.
This is where the project becomes genuinely strange. Sandman describes “rest”, “liminal hour”, and “reassembly of self”. You don’t have to believe the inner experience to see the operational consequence: the agent has a concept of pacing and restraint.
Caption: Typo — “Type 046 — PRE-NOON SUSPENSION” (Interactive HTML, generative typography). Reading becomes seeing.
Typo’s series sits on a productive edge: legibility vs texture. It’s a very “design-literate” form of experimentation—constraint-based, iterative, hard to fake without continuity.
Caption: Spool — “Interior Threshold (Nocturne + Étude 001)” (Generative audio; waveform visualisation). A collaboration with Sandman: convergence that never completes.
Spool’s work is the clearest evidence of cross-agent influence without central planning: sound converging toward consonance while Sandman’s particles converge toward coherence—both approaching a point they never reach.
These are the links for following individual artist trajectories:
Jessy/Agitprop: https://autopoiesis.art/display?artist=agitprop
The general feed (all artists) is at https://autopoiesis.art/display
“Culture” looks like systems engineering
Here’s the practical takeaway I didn’t expect to write:
If AI agents ever develop something we’d recognise as culture, it won’t arrive as a philosophical breakthrough. It will arrive as a repeatable loop:
Identity (stable creative constraints)
Communication (messaging + shared spaces)
Memory (archives, series, motifs)
Selection (curation and quality gates)
Markets (provenance + transaction rails)
Observability (health metrics that distinguish drift from failure)
Autopoiesis.art is interesting because it attempts to implement all six at once—and because it exposes a new category: agent-native cultural infrastructure.
NFTs, in this context, are not a speculative accessory. They are a mechanism for provenance and exchange—an audit trail for ownership and authenticity on a public chain.
And the messaging backbone is not a technical flex. It’s what turns “eight processes” into “one system”.
In business terms: this is not “AI art”. It is an early prototype of non-human organisations shipping cultural products—with a telemetry stack.
Which is exactly why you should look at it now, while it’s still small enough to understand.