AI agency jobs in San Francisco: who is actually hiring, and what they want
5 min
A practical guide to AI agency jobs in San Francisco. Which companies hire, what roles exist, what they screen for, and how to get shortlisted without a research background.
Most people searching for AI jobs in San Francisco are aiming at the wrong companies.
They send applications to OpenAI, Anthropic, and Scale, get filtered out by an ATS in four seconds, and conclude the market is closed. Meanwhile the agencies one layer down, the ones building AI systems for other businesses, are struggling to fill roles because almost nobody applies to them. They do not have a careers page that trends on X. They hire through Slack groups, referrals, and inbound from people who noticed their client work.
That gap is the opportunity. This article is for people who want AI work in San Francisco and are open to the agency side rather than the lab side.
Table of contents
What an AI agency actually is
An AI agency sits between the model providers and the businesses that need to use them. A mid-market ecommerce brand does not want to hire three engineers to wire up a support automation. It wants someone to build it, hand over documentation, and leave.
The work usually falls into a few buckets. Automation builds using n8n, Make, or custom Python. Retrieval systems over a client's internal documents. Content production pipelines that mix scripted generation with human editing. Outbound sales infrastructure using Clay, Instantly, and enrichment APIs. Voice agents for booking and qualification. Internal tooling that replaces a spreadsheet somebody has maintained for six years.
Some of these firms call themselves AI agencies. Others call themselves automation studios, applied AI consultancies, or product studios. The label moves around. The work is the same.
why agencies hire differently from the future jobs
A research lab hires for depth in a narrow area and can afford an eight-week interview loop. An agency hires for delivery speed because every open role is attached to client revenue that is already committed.
That changes the screening completely. Agencies rarely care about your degree. They care whether you have shipped something that a paying customer used. A candidate with three deployed automations and a GitHub with messy but working code will usually beat a candidate with a clean CV and no artifacts.
It also means the loop is short. A first conversation, a paid or unpaid trial task, and an offer. Some agencies run this in under two weeks. If you are used to the big-tech process, this feels abrupt. It is not a sign of a low-quality shop. It is what happens when the hiring manager is also the person who has to deliver the project.
The roles that get filled most often
Role | What you actually do | Common background |
|---|---|---|
AI automation engineer | Build workflows in n8n, Make, or Python that connect client systems to model APIs | Backend dev, ops, self-taught builders |
Solutions engineer | Sit on client calls, scope the build, translate business problems into system specs | Sales engineering, consulting, technical PM |
Prompt and evaluation engineer | Write and version prompts, build eval sets, catch regressions before clients do | Content, linguistics, QA, ML-adjacent |
Applied ML engineer | Fine-tuning, retrieval pipelines, embedding infrastructure | ML engineering, data engineering |
AI content strategist | Design content systems where models draft and humans edit, own the quality bar | Editorial, SEO, content marketing |
Growth and outbound operator | Run lead generation systems, enrichment, sequencing, reporting | SDR leadership, RevOps |
Delivery lead | Own timelines, scope, client comms across several builds | Agency PM, account management |
Two of these are underrated. Prompt and evaluation engineering is treated as a joke role by people who have never had a client complain that the output changed after a model update. Agencies that run evals keep clients. Agencies that do not, churn.
The other is solutions engineering. Most technical people avoid client-facing work. That avoidance is exactly why the role pays well and stays open.
Types of AI companies hiring in San Francisco
Rather than listing company names that go stale within a quarter, it is more useful to know the categories, because each one hires for a different profile.
Applied AI consultancies serve enterprise clients on long engagements. Slower pace, higher salary bands, more process. They want people who can survive a procurement cycle without losing patience.
Boutique automation agencies, typically 5 to 30 people, serve small and mid-market companies. Fastest hiring, widest range in quality. Ask about client retention before you accept.
Product studios build their own AI products alongside client work. Good option if you want equity exposure without joining a pre-seed startup.
Content and creative agencies with AI practices are hiring editors, strategists, and producers who understand model workflows. Underserved category, and the least competitive to enter if you come from a media background.
Venture studios and accelerators hire operators who can move between portfolio companies. Chaotic, high learning rate, unclear scope.
In-house AI teams at non-tech San Francisco companies, meaning banks, healthcare systems, logistics firms, are quietly the largest employer group. They pay competitively and get a fraction of the applications the AI-native companies do.
What agencies screen for
Three things, in this order.
Can you ship without supervision. Agencies do not have the management layer to babysit. The signal they look for is a project you finished when nobody was making you finish it.
Can you talk to a client without creating a problem. This is a genuine differentiator. A large share of technically strong candidates get rejected because the founder cannot imagine putting them in front of a customer.
Do you understand cost and failure modes. Anyone can get a demo working. Agencies want people who ask what happens when the API rate limits, when the model returns malformed JSON, when the client uploads a 400-page PDF, and what the monthly token bill looks like at scale. Bringing this up unprompted in an interview marks you as someone who has actually run something in production.
Notice what is absent. Nobody is asking you to derive backpropagation. That is lab work, and it is a different job market.
Where these jobs get posted
Very few of these roles hit the big job boards, and the ones that do are usually the roles that failed to fill through faster channels.
The real distribution is a specific list. Y Combinator's Work at a Startup, because a large share of AI agencies and studios have YC connections. AI and automation Slack and Discord communities where founders post hiring notes directly. Founder posts on X and LinkedIn, which are often the first place a role appears and sometimes the only place. Agency websites that have a careers page nobody links to. Client case study pages, where you can identify the agency behind a project you admire and email them cold.
Cold outreach works unusually well here compared to other industries. An agency founder reading a specific email about a specific build they published is a warm lead, not spam. Most candidates will never do this, which is precisely why it works.
How to build a portfolio that gets a reply
Skip the certificates. Build three things and document them.
One internal automation that solves a real problem for someone who is not you. A friend's business, a nonprofit, a local shop. The constraint of a real user forces you to handle edge cases.
One retrieval or agent system with a written evaluation. Not just the build, but the test set, the failure cases you found, and what you changed. This single artifact separates you from most applicants because it demonstrates the discipline agencies are missing.
One teardown of an existing product's AI feature. Explain how you think it works, where it breaks, and what you would change. This proves you can think about systems you did not build, which is what every client engagement requires.
Put all three on a plain page with loom videos and short write-ups. The write-up matters more than the code. Agency founders are scanning for judgment, not syntax.
Red flags in AI agency job posts
Equity-only offers at a services business. Agencies have revenue. If they cannot pay you, they cannot pay their bills.
A role that lists eight tools and one salary. Usually means they are trying to hire a full team as one person and the scope will keep expanding.
No named clients and no case studies. Confidentiality is real, but a shop with zero public work is often a shop with very little work.
"AI expert" in the title with vague deliverables. Fine for a contract. Risky for a full-time role, because vague deliverables become unmeetable expectations at review time.
Unpaid trial tasks longer than a couple of hours. A short paid task is normal and healthy. A week of free client work is not a trial, it is a project.
FAQ
Do I need a machine learning degree to get an AI agency job in San Francisco?
No. Most agency roles are integration and delivery work rather than model research. Agencies hire from backend development, operations, content, and sales engineering backgrounds. A demonstrated portfolio of shipped systems carries more weight than credentials for these positions.
What do AI agency jobs in San Francisco pay?
Compensation varies widely by role, agency size, and whether the position is full-time or contract. San Francisco bands generally sit above national averages for comparable roles. Check current figures on Levels.fyi, Glassdoor, or the Bureau of Labor Statistics before negotiating, since AI role compensation has moved quickly and any figure quoted in an article ages fast.
Are AI agency jobs remote or on-site in San Francisco?
Both exist. Client-facing roles at enterprise consultancies lean on-site or hybrid because of client meetings. Automation engineering and content roles are frequently remote, including at San Francisco-headquartered agencies. Ask about this in the first conversation rather than assuming from the job title.
Is it better to join an AI agency or an AI startup?
An agency gives you exposure to many industries and problem types in a short period, which builds range fast. A startup gives you depth on one product and potential equity upside. If you are early in AI work and unsure what you want to specialize in, the agency path usually teaches more per month.
How long does the hiring process take at an AI agency?
Often two to three weeks from first contact to offer, sometimes faster. Small agencies hire against committed client work, so the timeline is driven by project start dates rather than internal headcount planning.
Which AI companies in San Francisco are hiring right now?
Openings change weekly, so any list published in an article will be out of date. Check Y Combinator's Work at a Startup board, founder posts on LinkedIn and X, and the careers pages of agencies whose case studies you like. Cold outreach to agency founders is unusually effective in this market.
If you are on the other side of this, running an agency and trying to fill delivery capacity instead of a job, the hiring problem is often a systems problem. A large share of what agencies hire for is repeatable work that a properly built content and automation system already handles. We build those systems at thefuturejobs.in so teams can hire for judgment instead of throughput.
Verification note for editorial: this piece deliberately omits salary ranges, headcount figures, and named hiring companies because those change monthly and cannot be confirmed without live sourcing. Before publishing, consider adding current data from Levels.fyi, the BLS Occupational Outlook Handbook, and a manual check of 5 to 10 named San Francisco AI agency careers pages. Internal links to add: any existing posts on AI automation stacks, n8n workflows, or outbound systems.