Can The 50-year-old Actor Model Rescue Agentic Ai?

2025 is shaping up to beryllium nan twelvemonth agentic AI crossed from investigation to production. Gartner reports that 34% of businesses now deploy AI agents, and by 2028, up to 15% of regular activity decisions will beryllium made autonomously, pinch a 3rd of endeavor package applications expected to see immoderate shape of agentic capability.

Yet truthful far, nan committedness down agentic AI hasn’t matched nan reality connected nan ground.

“Most agentic AI projects correct now are early shape experiments aliases impervious of concepts that are mostly driven by hype,” said Anushree Verma, a elder expert astatine Gartner. “This tin unsighted organizations to nan existent costs and complexity of deploying AI agents astatine scale, stalling projects from moving into production.”

The problem, arsenic it turns out, mightiness not beryllium nan intelligence of AI models that person nan spotlight — it’s astir apt their cloud infrastructure too.

Agentic AI introduces a caller benignant of workload: thousands aliases moreover millions of semi-autonomous processes that perceive, reason, enactment and collaborate complete time. They clasp memory, discourse and goals. These properties break astir of nan unreality architecture patterns that nan manufacture has spent 2 decades perfecting.

To understand why truthful galore agentic projects stall, and what mightiness unblock them, The New Stack said pinch Matthew Gregory, laminitis and CEO of Autonomy, a caller level arsenic a work (PaaS) built wholly astir what is called nan actor model. Having spent years exploring what it takes to tally autonomous, distributed systems astatine accumulation scale, Gregory and his squad are assured that this is really this adjacent activity of AI products will beryllium built.

“Familiar unreality architecture patterns person really go barriers,” Gregory said. “Agentic products usability otherwise than modern web apps, and they request devices designed for that difference.”

How nan Actor Model Provides a Foundation for AI Agents

The character exemplary isn’t new. First conceived successful 1973, it was inspired by concepts successful physics for illustration wide relativity and quantum mechanics. It has since go an elegant shape for managing concurrency, 1 that softly powers immoderate of today’s astir demanding systems, including WhatsApp, Discord and Databricks.

In nan character model, nan basal portion of computation isn’t a instrumentality aliases function, but an character — a lightweight, independent entity that “owns” its ain state, processes messages asynchronously and communicates pinch different actors done connection passing. Hundreds of thousands of actors tin tally successful parallel wrong a azygous process: starting up successful milliseconds, idling cheaply and persisting for arsenic agelong arsenic they are needed.

“If you put an supplier successful a container, it idles while you salary for nan capacity. At scale, that’s a DevOps nightmare.”
— Matthew Gregory, laminitis and CEO of Autonomy

For AI agents, modeling each 1 arsenic a azygous character is an almost cleanable fit. Actors (agents) are:

• Stateful: Each character remembers its identity, memories, discourse and goals.
• Long-lived: Actors tin enactment idle for agelong periods without wasting resources.
• Asynchronous: They output gracefully while waiting for large connection exemplary (LLM) completions aliases instrumentality responses.
• Massively parallel: Millions tin coordinate and collaborate to support agentic patterns for illustration subagents and orchestrated workflows.
• Built-in messaging: Mailboxes and queues make unafraid connection natural.

“An character has a mini footprint and handles concurrency and messaging beautifully,” Gregory explained. “It’s nan correct atomic portion for agentic systems.”

Why Containers and Serverless Fall Short

Simply put, accepted unreality primitives weren’t designed for agentic workloads. While serverless functions are cleanable for short, stateless bursts of activity that are measured successful milliseconds, they are overmuch little suitable for an supplier that must clasp representation and discourse crossed long-running tasks. Reconstructing an agent’s authorities pinch each invocation adds latency and cost, while besides breaking continuity.

Containers, connected nan different hand, sphere authorities but travel pinch weight. Keeping thousands — fto unsocial millions — of mostly idle containers online is operationally achy and financially unsustainable.

“If you put an supplier successful a container, it idles while you salary for nan capacity. At scale, that’s a DevOps nightmare,” Gregory said. “Go serverless, and nan supplier forgets who it was each clip it wakes up. Actors springiness you persistent authorities and exceptional costs efficiency.”

In different words, actors fto developers deliberation successful position of relationships and collaboration without nan operational overhead aliases assets discarded that causes astir agentic prototypes to fail.

Solving Trust and Data Security for Autonomous Agents

Even if nan issues of concurrency and standard are solved, different situation quickly surfaces: trust. Agents don’t conscionable compute; they must besides pass — often crossed clouds, services and organizations. Each relationship must beryllium authenticated, encrypted and verifiable.

Autonomy addresses this pinch Private Links, a unafraid messaging furniture built connected apical of nan unfastened root Ockam toolkit that Gregory’s squad launched respective years ago. Every supplier connected nan level mints a unsocial cryptographic non-human personality (NHI), past establishes mutually authenticated, end-to-end encrypted channels pinch its peers. This eliminates immoderate request for VPNs, nationalist endpoints and shared secrets successful nan cryptographic sense.

“Data successful mobility is 1 of nan hardest problems for agentic developers,” Gregory said. “Private Links lick that pinch opinionated cryptography and protocols baked successful from Day 1. Better yet, an supplier moving connected nan Autonomy Computer tin create a unafraid relationship pinch immoderate different agent, exemplary discourse protocol server, tool, exertion aliases information shop successful immoderate different unreality aliases network.”

That practice matters. Ockam — an identity, messaging and access-control room pinch a vibrant unfastened root organization — has been battle-tested for unafraid connection betwixt distributed systems. Autonomy extends those aforesaid spot primitives to nan world of AI agents.

“In a way, we’ve sewage a five-year caput commencement connected nan problems that request solving to vessel agentic products,” Gregory noted. “Autonomy builds straight connected that foundation.”

From Pieces to Platform

Even pinch nan correct architecture and information model, galore teams still drown successful integration work. The emblematic agentic stack coming involves a twelve frameworks — LangChain for LLM tooling, CrewAI for multiagent coordination, positive vector databases, schedulers, observability layers, identity, possibly moreover immoderate Kafka and a messiness of civilization glue. Each solves 1 piece; nary shape a coherent whole.

“Infrastructure is difficult to standard and easy to get wrong,” Gregory said. “As a merchandise company, you extremity up spending each your clip connected plumbing alternatively of product.”

Autonomy’s level arsenic a work intends to adjacent that gap. It packages nan character runtime, spot furniture and observability devices into a azygous environment. Developers build locally pinch their Python SDK, past deploy into a globally distributed runtime without rubbing containers, Kubernetes aliases serverless orchestration. Scaling, routing and unafraid messaging travel built in.

At first glance, Autonomy mightiness look for illustration different agentic framework, but location is an important distinction. Frameworks thief you build; platforms thief you ship. Gregory illustrates it good pinch this analogy: “It’s for illustration nan narration betwixt Next.js and Vercel, aliases betwixt Rails and Heroku. Frameworks get you to code; platforms get you to customers.”

“Just telling group to ‘go usage actors’ doesn’t make each nan problems of distributed systems spell distant — it conscionable hands them a caller group of difficult engineering challenges.”
— Matthew Gregory

Actors Need Good Architecture Too

While nan character exemplary offers nan correct instauration for agentic systems, it isn’t a metallic slug conscionable by itself. Engineers who person antecedently built actor-based frameworks will cognize nan trade-offs well: tracing a azygous petition crossed thousands of asynchronous actors tin beryllium notoriously difficult; unbounded concurrency tin lead to connection storms; and recovering authorities if an character crashes aliases restarts requires observant orchestration. The character exemplary useful awesome for agents, but it besides needs immoderate guardrails too.

“Those are each existent problems,” acknowledged Gregory. “Just telling group to ‘go usage actors’ doesn’t make each nan problems of distributed systems spell distant — it conscionable hands them a caller group of difficult engineering challenges. Building a runtime that handles supplier identity, tracing, scheduling, nonaccomplishment betterment and information correctly is simply a serious portion of infrastructure work. That’s precisely nan load we wanted Autonomy to return disconnected nan developer’s plate.”

By recognizing nan long-known challenges pinch actor-based runtimes, and by designing its level to reside them, Autonomy turns a once-esoteric architecture into thing that thin teams tin usage to vessel reliable, production-grade agentic systems.

Developer Experience: Small Teams, Big Ambitions

Currently, Autonomy’s first assemblage is small, fast-moving teams — startups and “startup-mode” groups wrong larger enterprises that request to vessel quickly but who are hamstrung by a constricted headcount. These teams can’t spend to build bespoke infrastructure that is tuned to standard earlier they vessel to production.

“We’re really, really obsessed pinch developer experiences,” Gregory said. “Teams constrained by people, clip aliases resources will request thing that conscionable works, thing that’s automagical, astatine afloat scale.”

The level already includes enterprise-grade devices specified arsenic logs, metrics, evals and unafraid endpoints for observability and operations. But notably, Autonomy’s biggest advantage seems to beryllium its simplicity. Developers tin attraction connected what their agentic products tin do, not really to make them run.

Why This Architectural Shift Matters Now

If nan past decade of unreality computing was astir elastic compute, nan adjacent 1 will beryllium astir elastic autonomy — moving not conscionable much servers, but much decisions. The architectural portion of that early isn’t a instrumentality aliases function, it’s an actor.

The character exemplary provides nan instauration for systems that tin think, retrieve and coordinate astatine scale. It’s an thought that has existed for half a century, but nan caller emergence of agentic AI yet gives it a renewed mainstream purpose.

Autonomy’s stake is that a PaaS built astir actors — pinch spot and information woven successful — will go nan default substrate for these systems. Teams that clasp this exemplary won’t conscionable build smarter agents; they will vessel production-ready products faster.

“We’re not inventing a caller thought truthful overmuch arsenic putting a 1000 architectural decisions into a azygous level and offering it arsenic a elemental merchandise experience,” Gregory said. “Actors person ever fresh this problem. Now, nan Autonomy level makes it easy for each builder to vessel a scaleable, connected agentic product.”

Group Created pinch Sketch.