Anthropic Pays $850K for AI Researchers. Primitive Is Paying $320K for Engineers Who Build the Nervous System Between Agents.

What Primitive Is Actually Building

The AI industry has no shortage of startups chasing a better foundation model. Primitive is not one of them.

The San Francisco company is building the communication layer for AI agents: the infrastructure that lets independent agents exchange messages, coordinate tasks, and share state in real time. Think of it as the messaging backbone for multi-agent systems: not the brain, but the nervous system. That distinction shapes every hiring decision the company makes.

Most agent frameworks (LangChain, OpenAI Agents SDK, AutoGen, CrewAI) define agents around four core primitives: a reasoning model, tools and actions, per-agent memory, and orchestration logic. These frameworks handle what a single agent needs to function. They don't handle what happens when ten agents need to agree on what "revenue" means, or when a fleet of agents must coordinate without contradicting each other. Atlan's analysis found that agents built on four primitives alone show a 60% proof-of-concept abandonment rate before reaching production. The missing pieces are a shared context layer and a control plane (what Atlan calls Primitives 5 and 6).

Primitive sits in the gap those frameworks leave open. Its infrastructure handles agent-to-agent communication, the routing and delivery of messages between autonomous agents operating across different tasks and environments. This is a distributed-systems problem, not a model problem. The hard part isn't getting an LLM to generate a response — it's guaranteeing that response reaches the right agent, in order, without duplication or loss, at scale.

That's why Primitive's founding-engineer roles read like a job post for a real-time messaging platform, not an AI lab. The requirements (Docker, Kubernetes, infrastructure-as-code, CI/CD pipelines, Argo) are the tooling of engineers who've built and operated high-throughput, low-latency systems. The company isn't hiring people who fine-tune models. It's hiring people who've kept critical systems running under load.

As Praveen Gollamandala outlined in his map of agentic AI architectures, the field has four layers: LLM foundations, single-agent capabilities, multi-agent coordination, and an operational ecosystem for governance and reliability. Most venture capital and talent attention has concentrated on the first two. Primitive is building in Layer 3 — the coordination and communication infrastructure that has to exist before Layer 4's governance and observability tools have anything to observe.

The company's hiring spree is a signal about where the bottleneck is moving. The scarce talent isn't researchers who can squeeze another benchmark point out of a transformer. It's engineers who can build the communication substrate that makes multi-agent systems work outside a demo.

Decoding the Founding-Engineer Job Posts

Primitive's open listing for a Founding Infrastructure Engineer reads less like a standard job post and more like an architecture spec. The requirements are specific enough to reveal what the company is actually building — and what it isn't.

The non-negotiable toolkit: Docker, Kubernetes, infrastructure-as-code, CI/CD, and Argo. That last one matters. Argo is a GitOps continuous delivery engine for Kubernetes that treats cluster state as version-controlled declarations rather than imperative commands. A Medium post on self-hosting Argo CD describes using it to "prepare a k8s cluster fully automated," which is exactly the pattern Primitive seems to be betting on: infrastructure that configures itself from a repo, with no manual steps between commit and production.

The post asks for someone who has "built massive distributed systems from 0 to 1" and then scaled them. That's a narrow profile: not a machine learning engineer who picked up Terraform on the side, but someone who has owned the full lifecycle of a distributed system (designed it, shipped it, and kept it running when traffic hit).

There's also a line that stands out from most infrastructure listings: "Should be one of the most AI enabled engineers in the world. You should have at least 2 agents running right now." Primitive offers an unlimited AI budget for all team members, and the post frames that as a recruiting hook. The signal is twofold. First, Primitive wants engineers who already use AI agents in their daily workflow — not as a novelty, but as a core part of how they work. Second, the company is building exactly that kind of agent communication infrastructure, so having engineers who are heavy agent users means they're dogfooding the exact problem Primitive is solving.

The "etc." after the tool list is doing real work. The post says "Docker, k8s, IaC, CI/CD, Argo, etc. Even better if you have opinions on why this list is wrong." That's an invitation to engineers who don't just operate these tools but have strong views on their tradeoffs. For a four-person team at pre-seed stage, that kind of opinionated generalist is the only hire that makes sense.

Compensation sits at $240K to $320K base plus 2–4% equity, aggressive for a pre-seed company this size. But the skill set they're targeting commands that range in the market, and Primitive is competing against well-funded AI labs for the same infrastructure engineers. The salary band is the clearest signal that the company knows exactly how scarce this profile is.

Why Agentic Infrastructure Became the Hottest Hiring Frontier

Primitive isn't hiring in a vacuum. The company is entering a lane (agent communication, orchestration, and infrastructure) that has become the most competitive hiring surface in AI, precisely because capital has only started to catch up to where engineers are needed.

U.S. AI startups raising $100 million or more didn't slow in 2025 compared to 2024, per TechCrunch's roundup of 55 companies. But the distribution of that capital is lopsided. A separate analysis of the top 25 AI agent companies shows they pulled in more than $25 billion across 2025 and early 2026, with roughly 63% flowing to foundation model labs, 22% to infrastructure and tooling, and only about 15% to application-layer companies. That gap between where the money concentrates and where the actual deployment bottlenecks live is exactly where companies like Primitive operate — and why they can demand scarce talent.

The application layer's constraints are visible in the numbers. McKinsey reported that only 1% of companies describe their AI rollouts as mature, even as 78% of professionals plan to implement agents. The shortfall isn't model capability. It's infrastructure: 86% of enterprises need tech stack upgrades to deploy agents effectively, research from DevSquad found. That demand has turned agentic infrastructure roles into the ones companies fight hardest for.

The specific sub-sector Primitive sits in (agent-to-agent communication and interoperability) is still early but attracting serious attention. Google's Agent2Agent (A2A) protocol, launched in April 2025 with backing from more than 50 technology partners, is the most significant standardization push in agent communication to date. CopilotKit, a Seattle startup building technology to embed AI agents inside existing software applications, raised $27 million with participation from Google, Microsoft, Amazon, and Oracle, GeekWire reported. These are infrastructure plays, not model plays, and they signal that the hiring competition extends well beyond the foundation labs.

The talent market reflects this pressure. The San Francisco Bay Area accounts for over 31% of AI jobs and draws nearly one-third of AI/ML professionals, LinkedIn's 2025 talent data shows. CBRE's Scoring Tech Talent report found that tech workers with AI skills in the Bay Area grew 24% year over year in 2025. That supply is still nowhere near demand when the roles require real-time distributed systems experience, container orchestration at production scale, and the kind of reliability engineering that agent-to-agent communication demands — the exact stack Primitive's job posts list.

What makes this hiring frontier different from the AI talent wars of 2023 and 2024 is the profile of the engineer. Then, the fight was over ML researchers and training infrastructure specialists. Now, the bottleneck has shifted to the layer between the model and the deployment: the people who can build what the funding map calls "infrastructure and tooling" and what enterprises actually experience as the thing standing between a prototype agent and a production one.

The Compensation Signal: What Primitive Is Paying to Win

Primitive hasn't published its salary bands publicly, so the exact numbers it's offering founding engineers remain opaque. But the company's hiring context (San Francisco, early-stage, infrastructure-first) lets us triangulate what it almost certainly needs to pay.

The benchmark is right next door. Anthropic's current open roles on Zero G Talent's board list total comp ranging from $350,000 to $850,000 per year for research and staff engineering roles in San Francisco. OpenAI's board listings show a similar band, with senior technical roles reaching $295,000 to $555,000. Those are well-funded, later-stage companies with name recognition and (in Anthropic's case) a product in heavy use. Primitive has none of that.

For an early-stage startup competing for the same pool — engineers who can design real-time distributed systems, build agent-to-agent protocols, and operate infrastructure at scale — the math is unforgiving. Pre-product, pre-revenue companies in SF typically need to offer either equity-heavy packages with moderate base salaries or match late-stage cash comp and bet that the equity story closes the gap. Primitive's job posts emphasize infrastructure ownership and early-team influence, which is the standard pitch. But the engineers they're targeting have offers from OpenAI and Anthropic on the table.

The broader market data backs this up. Founding infrastructure engineers at well-funded AI startups in the 2024–2025 cycle have been commanding base salaries of $200,000–$300,000 plus meaningful equity grants, according to levels.fyi and reported offer data. At the top end (staff-plus roles at companies with strong investor backing) total comp can push past $500,000 when equity is valued at preferred price. Primitive's roles, which ask for hands-on ownership of distributed-systems architecture from day one, sit squarely in that range.

What's telling is what Primitive's job posts don't emphasize. There's no mention of large team sizes, management tracks, or established product-market fit. The implicit offer is technical scope and proximity to the problem. That works for a certain kind of engineer: the type who turned down a $400,000 staff role to build something from scratch. But it also means Primitive can't win on comp alone against Anthropic's current bands. It has to win on the problem, the team, and the equity upside.

For engineers evaluating agentic-infrastructure roles more broadly, the market for distributed-systems talent in AI infrastructure is tight and getting tighter. Companies like Primitive are paying a premium not just for the skill set itself but for the willingness to apply it to a problem (agent communication) that doesn't have established patterns yet. The engineers who can define those patterns are, right now, among the most expensive hires in tech.

The Founder's Pedigree: What the Resume Reveals

Primitive's Y Combinator profile is unusually terse about the founding team's pedigree, but the signal is clear: "Formerly founder of Actual AI and engineering @ Microsoft/AWS/FB/Google." That single line compresses a career arc that tells you almost everything about what Primitive is likely building and how it plans to build it.

The founder-of-Actual-AI detail matters. Actual AI was a startup in the applied-AI space, the kind of company that ships product, not papers. Someone who went through the zero-to-one cycle once already, then rotated through engineering roles at four of the largest infrastructure operators on the planet, is not starting a research lab. They're starting a systems company. The sequence (founder, then Microsoft, Amazon, Meta, Google) reads like a deliberate tour through every major cloud and distributed-systems stack in the industry. Each of those companies runs real-time messaging and agent-adjacent infrastructure at a scale most startups never approach.

That background maps directly onto the job posts. When Primitive's founding-engineer listings call for Docker, Kubernetes, Argo, and real-time distributed systems, they're not describing a generic devops wish list. They're describing the exact stack you'd spec if your founder had spent years inside the teams that built and operated those systems at hyperscale.

The broader YC founder-genealogy data from FounderTrace reinforces the pattern. Y Combinator's network maps show that infrastructure-focused startups tend to cluster around founders who came from other infrastructure companies — not from model labs or consumer apps. Primitive fits that mold. Its founder didn't come from a frontier-AI research group. They came from the layer below it: the compute, the orchestration, the pipes.

Primitive is unlikely to compete on model quality or agent reasoning. It's positioning where the founder has the deepest operational knowledge — the communication and coordination layer that lets agents talk to each other reliably, at scale, in real time. The hiring spree for engineers with production distributed-systems experience, rather than ML researchers, confirms it. The team is being built to ship infrastructure, not to publish benchmarks.

Every AI Lab Will Need This Layer

Primitive's job posts are a leading indicator of a shift most AI labs haven't fully reckoned with. The industry's talent bottleneck is moving — from model training to the infrastructure that lets agents actually work together in production.

McKinsey's 2025 Global Survey on AI found that 23% of respondents report their organizations are already scaling at least one agentic AI system. Sixty-two percent are experimenting with agents in some form. But the same data reveals the constraint: in any given business function, no more than 10% of respondents say they're scaling agents. Most are stuck in pilots. The gap between experimentation and production is where infrastructure (not model capability) becomes the bottleneck.

That gap is exactly what Primitive is hiring to close. When a company lists Kubernetes, Argo, real-time distributed systems, and CI/CD pipelines as non-negotiable requirements for founding engineers, it's signaling that the hard problem isn't building the agent. It's building the layer that lets agents talk to each other reliably, at scale, under real production load.

Enterprise AI adoption is broadening: 88% of McKinsey respondents report regular AI use in at least one function, up from 78% a year earlier. AI high performers, the roughly 6% of organizations attributing meaningful EBIT impact to AI, are at least three times more likely than peers to be scaling agents across functions. Those high performers also share a pattern: they've redesigned workflows, invested in data infrastructure, and built the platforms needed to run agents in production rather than in demo environments.

Meanwhile, the talent market is straining in a new direction. McKinsey's survey found that software engineers and data engineers are the most in-demand AI-related roles, with larger companies (over $1 billion in revenue) roughly twice as likely as smaller firms to be hiring for roles that integrate, model, and industrialize data. Deloitte's earlier research on AI talent shortages found that even mature adopters reported major or extreme skill gaps at higher rates than less mature organizations, because working with AI more extensively reveals what skills are actually needed versus what leaders think they need.

As more organizations push agents from pilot to production, the scarce skill set shifts from training models to operating them. Real-time messaging between agents, reliability under load, observability, failure handling — these are distributed-systems problems, not machine-learning problems. Primitive's hiring spree is one early signal of that transition. The companies that solve the communication layer first won't just ship faster. They'll define the infrastructure standard that every other AI lab ends up building on.

What Engineers Should Take Away — and Where to Apply

The skill set Primitive is hunting for is specific and hard to fake. Docker, Kubernetes, infrastructure-as-code, CI/CD pipelines, Argo, real-time distributed systems — these aren't buzzwords on a job post. They're the actual substrate of agentic infrastructure. If you've spent years keeping backend services alive under load, you're closer to this wave than you think.

What to build, in priority order:

• Distributed systems fluency. Consensus, partitioning, failure modes, observability. If you can reason about what happens when a message drops between two agents mid-task, you're useful.
• Kubernetes and container orchestration. Primitive's stack leans heavily here. Know how to write operators, manage rollouts, debug networking issues between pods.
• Infrastructure as code and CI/CD. Argo, Terraform, Helm — the tooling that lets a ten-person team run a system that would've needed a platform group three years ago.
• Agent-to-agent protocol intuition. MCP, A2A, or whatever standard emerges. Understand the problem: agents need to discover each other, authenticate, hand off tasks, and recover from failure without a human in the loop.

Where the jobs are right now:

LinkedIn lists 565 AI agent infrastructure roles in the U.S. alone, with 13 posted in the past day. The spread is wide — founding engineer roles at well-funded startups, staff agent engineer positions at companies like Zuma, platform roles at ByteDance and Netflix, and early-stage ops hires at companies like Tessera Labs and Vapi.

How to actually get hired:

Seed and Series A startups filling these roles often skip recruiters entirely. GetClera reports that direct outreach to founders, referrals from Slack and Discord communities, and applications through AngelList (Wellfound) and YC's Work at a Startup page reach candidates that traditional pipelines miss. A portfolio project that shows you can deploy and observe a multi-agent system on Kubernetes will outperform a generic cover letter every time.

The bottleneck in AI is shifting from model training to deployment. Primitive's hiring spree is one data point. The 565 open roles are another. The engineers who can build that inter-agent communication layer (and keep it running) are the ones this market is bidding on.

Working in AI? Zero G Talent tracks the openings: browse AI jobs, openings at OpenAI and Anthropic, and the people building the field.