Bear with me now…the Fermi paradox. Given the number of stars, planets and opportunities for intelligent life to emerge, Enrico Fermi reportedly asked a simple question: where is everybody? Why don't we see evidence of civilisation wherever we look?
The ad industry seems to have developed its own version of the problem. Walk around Cannes, Possible or almost any industry conference and it's a barrage of agentic platforms, autonomous workflows and intelligent assistants. There are plenty of acronyms to connect them all as well.
The conditions ought to be favourable. Programmatic contains millions of repetitive decisions across planning, activation, optimisation, measurement and reporting. It has vast amounts of data, software-based workflows and constant pressure to reduce cost.
Building the basic technology is increasingly accessible too. I spent a couple of evenings creating skills, connecting tools through MCP and building agents in Codex and Claude. These weren't production systems, obviously, but they worked well enough to show how low the barrier to experimentation has become.
With so many possible applications and a relatively low barrier to building something, I expected to see more agents doing consequential work. Most of what I can actually see still looks like assistance with a more ambitious label.
Assistance and agency
A chatbot that answers a campaign question can save time, and a tool that recommends an audience may improve a decision. However, neither has much authority; there is still a person on the hook to make sure the selection of inventory or choice of partner can be explained to a client or boss.
Genuine agency starts when a company allows the system to make one of those choices without waiting for approval. The system can now spend money and produce an outcome, although healthy debate will rage over what qualifies as an outcome and how it will be attributed. Some things won't change, so let's welcome the 22nd-century equivalent of “last click wins”.
Many inhabitants of the agentic universe are assistants wearing slightly more impressive name badges.
The great filter
One explanation for the original Fermi paradox is the great filter, suggesting there's a stage that very few civilisations manage to pass, presumably because they destroy themselves with the very technologies they have built. Advertising agents may have a filter between an impressive demonstration and dependable production use.
The awkward work starts once the demo ends. Legal will demand a view on liability, finance needs an audit trail and the technology team has to make sure every connection still works. A client will also want a coherent explanation when an automated decision produces an unexpected result.
I suspect this is one reason conference demonstrations look so cheerful. Nobody in the demo has to explain an unexplained budget move to a client on Monday morning.
Protocols can help systems communicate and make connections easier to build. They can't decide whether a chief financial officer, client or regulator will trust the result. That decision belongs to the organisation, and many organisations haven't yet worked out how much authority they're comfortable handing over.
Quiet agents
Some of the most effective agents won't be announced. A company that has found a real advantage in planning or optimisation has little reason to explain how it works. The agent can sit inside an existing workflow with no launch event and no person-shaped avatar.
The evidence is more likely to appear in operating performance. A useful agent should make campaigns launch faster, allow each person to manage more activity, improve decisions, lower the cost of execution or make previously uneconomic work worthwhile. None of that requires a new interface.
If an agent works properly, the customer should only notice that performance has improved. How the system works is of no consequence. This is already true in Meta and Google today.
Some ideas don't survive contact with the economics
There's also a less exciting explanation for the limited deployment. Some proposed applications don't create enough value to justify the work required to make them dependable.
Calling a workflow agentic doesn't improve its economics because clients pay for outcomes, not workflows. The current enthusiasm has produced clever solutions searching for decisions they should be allowed to make. A buyer still has to trust the system with money and value the result enough to pay for it.
Where is everybody?
Most agents remain constrained or hidden. Many are useful assistants presented as autonomous systems. Others are experiments waiting for trusted data and clear permission, while the genuinely useful deployments may be operating quietly inside existing products.
The practical questions concern the decisions an agent can make and the rules it must follow. Companies also need to know who can stop it, how quickly an error can be reversed and who takes responsibility when the system gets a decision wrong. Until those points are settled, most will keep a person close to the final decision.
Most of what we're calling agentic today sits somewhere between an assistant and an autonomous system. Companies will grant more authority, one decision at a time. Until then, the agentic universe will continue to look strangely quiet.