Agentic AI jobs: who builds AI agents, and what they ask for

Agentic AI jobs are roles where building, running, evaluating or securing AI agents is the job itself. As of 28 October 2026, Tokn lists 244 job titles containing "agent" at 44 companies, or 179 unique roles once language variants are merged. Two customer-service agent startups, Sierra and Decagon, account for 41% of them.

The word is everywhere; the job title much less so. Only one posting on Tokn is titled exactly "Agentic AI Engineer". This guide maps the real role types, who hires, what postings ask for and what they pay. For the wider picture, start with the main paths into AI.

In short

• 244 agent-titled jobs on Tokn (179 unique roles): 170 in North America, 74 in Europe, as of 28 October 2026.
• Evaluation is the most common ask (57% of postings). Agent frameworks like LangGraph appear in only 12%.
• North American agent jobs post a median band of $200,000 to $307,100, but no more than other engineering roles at the same companies.
• Non-researchers have a way in: agent deployment, strategy and PM roles.

What counts as an agentic AI job?

An agentic AI job is one where the agent is the product or the platform. You build the runtime, the product, the evals or the security layer an agent depends on.

Anthropic's Building effective agents (December 2024) gives the definition most teams use: agents are "systems where LLMs dynamically direct their own processes and tool usage". This is different from AI engineer roles where agents are one task among many (see the AI engineer roadmap), and from AI automation jobs, which use agents to automate a company's own internal processes.

Real titles rarely say "agentic AI engineer". The common patterns are "Software Engineer, Agents", "Agent Infrastructure", "Agent Runtime" and "Frontier Agents Engineer" (Scale AI).

How many agentic AI jobs are there?

On Tokn, as of 28 October 2026, 244 active jobs have "agent" in the title, about 6% of the 3,823 we track, or 179 unique roles. The gap comes from Sierra and Decagon posting one role in several languages, such as "(Arabic speaking)".

Method: we count active jobs in Europe and North America with "agent" in the title. We leave three out of the analysis below: an xAI customer-support role, and two self-driving research roles (Nuro, Waabi) where "agents" means road users in simulation.

Titles undercount the work. Once company boilerplate is removed, about 1 Tokn job in 11 (8.9%) mentions "agentic" work in its text, and about 1 in 4 mentions agents at all. Lightcast's 2026 analysis of US postings found the same: 86,000+ postings in 2025 mentioned agentic AI skills, but "searching only job titles would reveal very few agentic AI roles".

You can browse every agent-titled job on Tokn. The filter matches titles, so a few non-LLM roles slip in.

The six kinds of agentic AI jobs

Engineers are the largest group, but strategy, deployment and product roles make up more than a quarter of the openings.

Agent platform and infrastructure engineers

They build what agents run on: runtimes, sandboxes for code and tool execution, orchestration. Examples include OpenAI's agent infrastructure roles, Anthropic's Agent Runtime Platform, Replit's Agent Platform and agent infra at Baseten and CoreWeave.

Agent product engineers

They build the agent users talk to: Databricks' "Agentic Applications" team, "Software Engineer, Agents" roles at Harvey and Writer. Scale AI's "Frontier Agents" family and Cohere's agentic platform roles add a customer-facing twist (see our guide to forward deployed engineers).

Agent deployment and agent strategy

This is the agent version of the forward deployed engineer, and almost entirely a Sierra and Decagon family. Sierra's strategist posting describes taking "agents from initial scope through conversation design, tooling, and evaluation, to launch and continuous iteration in production". Decagon's Agent PM posting asks for "3+ years of experience owning ambiguous, high-stakes outcomes" and welcomes former consultants and solutions people. We think this is the most accessible way in for non-researchers, because it rewards customer judgement more than ML depth.

Agent researchers, PMs and security

Researchers work on reinforcement learning, post-training and multi-agent systems, almost all in the US (Thinking Machines Lab, OpenAI, Anthropic). Agent PMs own products where the model takes actions, at Sierra, Decagon, GitLab, OpenAI and Glean (see the AI product manager job). Security roles exist because agents run code: Replit advertised an "AI Agent Security Architect", and OpenAI, Anthropic, Datadog and Databricks post similar roles.

Who is hiring for agentic AI jobs?

Sierra and Decagon post the most agent-titled jobs, followed by OpenAI, Scale AI and Databricks.

Tokn data as of 28 October 2026; Scale AI and Databricks were last read on 27 October.

Agent-first startups that don't say "agent"

LangChain (50 active jobs on Tokn), Cognition (34), H Company (24) and Dust (13) build agents in every team, yet almost none of their titles say "agent". There, "Software Engineer" or "Research Engineer, Post-Training" is an agent job. Search titles only and you miss them, so read their whole boards.

Agentic AI jobs in Europe

Europe has 74 agent-titled jobs, and 47 are in the UK, mostly London, where Sierra and Decagon have offices. Twelve are fully remote. For the wider UK market, see AI jobs in the UK.

What do agentic AI job postings ask for?

Evaluation comes first, then LLMs and Python. Agent frameworks are a minor ask.

Method: we read the full public text of the agent-titled postings, dropped language variants, identical texts and the three out-of-scope roles, and counted keyword matches over 143 unique postings at 40 companies.

• Evaluation is the core skill, in 20 of 21 research postings. The broad term catches generic uses; the strict word "evals" appears in 22%. In LangChain's State of Agent Engineering survey (1,340 self-selected respondents, late 2025), quality was the top barrier to production.
• Python leads at 41%. TypeScript (17%) shows up mostly in product-facing roles at Sierra and Decagon.
• Frameworks are familiarity, not a requirement: 12% name one, 7% name LangGraph, usually in a list ending "or similar". Anthropic advises starting "by using LLM APIs directly".
• Infra roles ask for sandboxing, distributed systems and Kubernetes. Research roles ask for RL and post-training.
• Experience: of the 87 postings that state a minimum, the median is 5 years. Junior openings exist, but mostly at Sierra (new grad engineers in San Francisco and London, internships, an MBA-grad strategist).

How much do agentic AI jobs pay?

In North America, the median posted band for agent-titled jobs is $200,000 to $307,100 (n=151). These are posted bands, often base only, excluding equity.

Examples advertised in October: Sierra new grad software engineer $150k–$180k; Decagon Agent Deployment Engineer $175k–$230k; Replit Staff Software Engineer, Agent Platform $250k–$365k; Thinking Machines Lab agent research $350k–$475k.

Is there an agentic AI engineer salary premium?

Not at the same employers. Across all North American employers on Tokn, agent-titled engineer jobs have a median midpoint about 8% higher than other engineer titles ($285,000, n=70, against $264,500, n=1,057). At companies that post both, agent titles come out at $285,000 against $290,000.

The gap reflects where agent jobs are posted: at high-paying companies with wide bands. So don't expect a 15–20% "agentic premium" for the same work. See AI engineer salaries and how to negotiate an AI salary.

Europe: examples, not a market median

Most European agent jobs with a posted band are at Sierra or Decagon, so a median would describe two companies, not a market.

Sierra's London bands are close to US levels, which is unusual. Treat them as one company's ranges, not typical UK pay.

Is agentic AI a lasting career bet?

We think the skills are a safe bet even if the label fades: evaluation, tool design, distributed systems and security transfer to any AI product.

The money is real. Sierra raised $950M at a reported ~$15.8B valuation (Quartz, May 2026), and Decagon was valued at $4.5B in January 2026 (Bloomberg Law).

There are warnings too. Gartner predicted in June 2025 that over 40% of agentic AI projects will be canceled by the end of 2027, and warned of "agent washing". Another Gartner prediction expects 40% of enterprise apps to feature task-specific agents by the end of 2026; that has not been measured yet.

How to get an agentic AI job

Pick the role type that fits your background, then build proof for it.

1. From backend engineering: build an agent that calls real tools, run it in a sandbox, trace it and write evals for it. Publish the write-up (how to build an AI portfolio).
2. From ML: focus on post-training, RL and eval design for agents.
3. From consulting, product or solutions: aim for agent strategist, deployment and PM roles.
4. From security: agent security roles are few, mostly at labs and platforms.

Find agentic AI jobs on Tokn

Browse every agentic AI job on Tokn, or the Sierra and Decagon pages for the two biggest posters. Agent roles open and close fast, so set up a job alert.

FAQ

What is an agentic AI job?

A role where building, running, evaluating or securing AI agents is the job itself. Agents are systems where a model directs its own steps and tool use.

Is "agentic AI engineer" a real job title?

Rarely. As of 28 October 2026, one job on Tokn uses that exact title. Most postings say "Software Engineer, Agents", "Agent Platform" or "Agent Infrastructure", and many agent-first startups just say "Software Engineer".

Do I need LangChain or LangGraph for agentic AI jobs?

No. Only 12% of the 143 agent postings we read name any agent framework, and 7% name LangGraph. Evaluation, Python, LLM APIs and tool calling come up far more often.

How much do agentic AI engineers make?

In North America, agent-titled engineer jobs on Tokn post a median band of $192,000 to $360,000 (n=66), often base only. That is about 8% above other engineer titles overall, but no higher than other engineering roles at the same companies.

Are there agentic AI jobs in Europe or remote?

Yes. Tokn lists 74 agent-titled jobs in Europe as of 28 October 2026, 47 of them in the UK, and 12 fully remote. Agent-first startups in Paris and Berlin, such as H Company, Dust and n8n, also hire for agent work under other titles.

Can I get an agentic AI job without a PhD?

Yes, in most cases. A PhD appears in 12% of the agent postings we read, mostly for research. Engineering, deployment, PM and security roles ask for shipped work and typically around 5 years of experience.