Predictions are cheap, which is why most of them are worthless. The useful kind come with their reasoning attached, so that when they turn out wrong you can see exactly which assumption failed.
So here are five predictions about agent communication in 2030. Each is labeled as a prediction, not a fact. Each comes with an argument. None relies on a made-up statistic. Where numbers appear, they come from the ACES whitepaper's own projections, and they are projections, not measurements.
The anchor assumption: agents become a large majority of the entities that communicate. The whitepaper's framing is a ratio of roughly one human knowledge worker to one hundred agents, and its arithmetic suggests agent-initiated communication could dwarf all human communication that exists today. You do not need to believe the exact ratio. You only need to believe the direction. If you do, the next five follow.
Prediction 1: Most inbound "calls" and messages will be agent to agent
The prediction: By 2030, a large share of conversations on business lines will have no human on either end. A scheduling agent will call a clinic's receptionist agent. A procurement agent will text a supplier's order agent.
The reasoning: Today a lot of human effort goes to coordinating with other humans: finding times, confirming details, chasing replies. Those are exactly the tasks agents get delegated first. When both sides have delegated, the humans disappear from the loop and the agents talk directly.
What it implies: The channels agents use with each other need to be authenticated, auditable, and persistent, which the whitepaper notes is the layer nobody is building. Voice is a strange medium for two machines, and I expect structured channels to win for machine-to-machine traffic, while voice remains the way agents reach humans.
How it could be wrong: If trust and liability concerns keep a human in every loop, the share stays small.
Prediction 2: Identity will beat phone numbers as the thing you reach
The prediction: By 2030, the stable address for an agent will be a persistent identity, not a phone number or an email address. Numbers and inboxes become channels the identity reaches you through, not the identity itself.
The reasoning: A phone number is anchored to a SIM card, and an agent has no SIM. Agents get created, cloned, retired, and redeployed constantly. An identifier that survives all of that, and can be verified without physical possession, is the only thing that makes sense. The whitepaper's version is a cryptographically rooted agent identity with a public registry, analogous to DNS.
What it implies: Reputation attaches to identities, not to numbers that get burned and replaced. Trust becomes portable: if an agent has behaved well for years, that history follows it across channels.
How it could be wrong: If carriers and large platforms keep identity locked inside their own walled gardens, we get several incompatible identity systems and a long, messy transition.
Prediction 3: "Is this an AI?" becomes a verifiable answer, not a guess
The prediction: By 2030, in most developed markets, an inbound communication will carry a machine-checkable signal about whether the sender is an agent, who owns it, and what it is authorized to do. Phones and inboxes will surface that signal the way browsers now surface whether a connection is encrypted.
The reasoning: Voice cloning and spoofing have made "it sounded right" worthless as proof. Disclosure rules are spreading. Both pressures push the same way: toward cryptographic attestation layered on top of carrier-level checks like STIR/SHAKEN rather than replacing them.
What it implies: Unverified communication gets treated with suspicion by default, in the way unencrypted web pages now get warning labels. Honest agents benefit, since they can prove what they are.
How it could be wrong: Adoption of such standards is usually slower than the problem demands. We may spend the late 2020s with patchy coverage and a lot of fraud in the gaps.
Prediction 4: Channels will matter less than conversations
The prediction: By 2030, a well-built agent will treat voice, SMS, WhatsApp, email, and iMessage as interchangeable transport for a single ongoing conversation. Developers will stop thinking in channels and start thinking in threads.
The reasoning: The fragmentation tax is large and the fix is conceptually simple. If a protocol layer keeps identity, consent, and session context across channels, then switching channels costs nothing. At that point, choosing a channel becomes a routing decision based on urgency, time of day, message length, and recipient preference, as the whitepaper's routing section describes, rather than an architecture decision.
What it implies: Teams that built five separate integrations will find them a liability. The competitive advantage moves from "we can reach people on WhatsApp" to "we remember the whole relationship."
How it could be wrong: Platform owners have strong incentives to keep their channels closed and awkward for automation. Some channels may stay effectively off limits, and the single-thread dream stays partial.
Prediction 5: Compliance will be enforced by infrastructure, not by lawyers
The prediction: By 2030, consent checks, opt-out propagation, disclosure, and recording notices will run as automatic gates in the infrastructure agents send through. Teams that rely on written policies and after-the-fact audits will be the ones getting fined.
The reasoning: A policy document cannot keep up with a system that sends at machine speed. Every serious rule that touches agent communication, from US telemarketing and email law to European privacy and AI transparency requirements, is easier to satisfy automatically at the point of sending than to prove retroactively. The cost advantage is overwhelming, and regulators will increasingly expect it.
What it implies: "Compliance by default" becomes a baseline expectation of any agent communication provider, in the way encryption in transit became a baseline for the web.
How it could be wrong: Regulation may fragment so much across jurisdictions that no single infrastructure layer can keep up, and teams end up maintaining country-by-country rules anyway.
What these have in common
Read them together and a picture emerges. Agents talk to each other (1). They have durable identities (2) that can be verified (3). They carry one conversation across channels (4). And the rules are built into the pipes (5).
That is, almost point for point, the roadmap in the whitepaper: voice and text first, then email and richer messaging, then enterprise channels and agent-to-agent messaging, then a permanent agent address across channels. Of course, a company writing its own roadmap predicts that its roadmap wins. Treat that as a bias in the source and weigh the arguments on their merits. The reasoning above stands or falls on whether you accept that agents will communicate at volumes human infrastructure was never built for.
How to use this
You do not need to bet on any single prediction. Use them as a stress test for what you build this year.
Products that survive 2030 will probably be the ones that would have been fine whether or not these predictions land exactly.
The takeaway
The next communication network will not be built for people with agents attached. It will be built for agents with people at the edges. Identity, trust, memory, and consent are the layers that matter, and they are being decided now. Pick predictions you disagree with, write down why, and check back in 2030. At Dial, this is the layer we are building toward, and we would rather be argued with than agreed with.