Skip to main content

Where to find work

In one line: In 2026 the highest-yield AI-engineering job search is two parts: a focused list of 15–30 named target companies, plus showing up in 2–3 communities long enough that opportunities come to you.

In plain English

The AI-engineering job market in 2026 does not look like the generalist SWE market. Most senior roles are filled via referrals or direct outreach from execs, not via job boards. Most junior roles are filled via portfolios + community presence + a small number of focused applications. Spray-and-pray on LinkedIn is a 2018 strategy; the 2026 strategy is "be a knowable person in a small number of places."

Where the jobs are (2026, named)​

Frontier labs​

Anthropic, OpenAI, Google DeepMind, Meta FAIR / GenAI, Mistral, xAI, Cohere, Reka, AI21, Adept (post-Amazon).

Highest bar. Highest comp. Slowest hiring loop (often 6–10 weeks). Mostly looking for senior engineers with shipped work; juniors get in via exceptional artifacts.

AI-native scaleups (the workhorse tier)​

  • Coding & dev tools: Cursor (Anysphere), Codeium / Windsurf, Cognition (Devin), Replit, Tabnine, Sourcegraph (Cody).
  • Search & retrieval: Perplexity, Glean, Hebbia, You.com, Exa.
  • Vertical AI agents: Harvey (legal), Sierra (CX), Decagon (CX), Cresta (CX), Mercor (recruiting), Crew, Lindy, MultiOn.
  • AI for finance & ops: Brex AI features, Ramp AI, Pry, Mosaic, Eilla, Rogo.
  • Voice AI: Vapi, Retell, Bland, PolyAI, Speak, ElevenLabs (Conversational), Hume.
  • Creative AI: Runway, Pika, Suno, Krea, Higgsfield, ElevenLabs, Cartesia.
  • AI infra / DevTools: Modal, Baseten, Together, Fireworks, Replicate, Anyscale, vLLM project, Braintrust, Langfuse, LangChain (incl. LangSmith), Vellum, Promptfoo, Inngest, Mastra, Trigger.dev, Restate.
  • AI productivity: Granola, Friend, Limitless, Mem, Otter.
  • Robotics + AI: Figure, 1X, Skild, Physical Intelligence, Tesla Optimus.

AI at big tech​

  • Microsoft AI (Copilot, Azure OpenAI, GitHub Copilot, M365 Copilot).
  • Google Cloud AI / Vertex AI / Gemini API.
  • AWS Bedrock / SageMaker.
  • Meta GenAI (Llama, Meta AI assistant, Ray-Ban Meta).
  • Apple Intelligence.
  • Amazon Q.

Stable, high comp, slower iteration, often less ownership per engineer than at a scaleup.

AI features at non-AI SaaS companies​

Almost every B2B SaaS has an AI team by 2026:

  • Productivity & docs: Notion AI, Linear AI, Atlassian Rovo, Asana AI, ClickUp Brain, Coda AI, Mem.
  • Design: Figma AI, Canva Magic, Adobe Firefly, Sketch AI.
  • CRM & sales: Salesforce Einstein, HubSpot Breeze, Gong, Outreach AI, Apollo AI.
  • Storage & search: Box AI, Dropbox Dash, Google Workspace Gemini.
  • DevOps & data: Datadog Bits AI, Snowflake Cortex, Databricks AI Functions, dbt Copilot.
  • Customer service: Intercom Fin, Zendesk AI, Freshworks Freddy.
  • HR & finance: Workday AI, Rippling AI, ServiceNow Now Assist, NetSuite SuiteAnalytics.
  • Consumer apps: Duolingo AI, Khan Academy Khanmigo, Headspace AI, Spotify DJ, Pinterest AI search, Reddit AI summaries.

Regulated industry AI teams​

  • Finance: Goldman Sachs internal AI, JPMorgan LLM Suite, Morgan Stanley AI @ Morgan Stanley, Two Sigma, Citadel, Bridgewater.
  • Healthcare: Hippocratic AI, Abridge, Suki, Nabla, Tennr, Glass Health, the AI teams at Epic, Cerner, UnitedHealth.
  • Legal: Harvey, EvenUp, Spellbook, Casetext (Thomson Reuters), the AI teams at major law firms.
  • Government: Anthropic / OpenAI government teams, Palantir, Scale AI federal.

Consulting & implementation​

Increasing demand for hands-on AI implementation help:

  • Pure AI consultancies: Decagon (also a product), Slalom AI, Slingshot, Eleos.
  • Big consulting firms' AI practices: McKinsey QuantumBlack, Accenture AI, BCG X, Deloitte AI Institute, Bain Vector.
  • Boutique: Wand AI, Synthesia for enterprise, custom Anthropic / OpenAI partner shops.

Where to actually look​

Job boards (real, not generic)​

  • Wellfound (formerly AngelList) — best for AI-tagged scaleup roles.
  • Y Combinator's Work at a Startup — high signal for early-stage AI companies.
  • Latent Space jobs board — curated, AI-focused.
  • AI Tinkerers job board.
  • Anthropic Builder Network jobs.
  • Hugging Face jobs.
  • GitHub Jobs (via individual repos).

Communities → jobs​

This is where most senior 2026 AI roles actually get filled:

  • AI Engineer Summit / AI Engineer World's Fair — the canonical AI-engineering conference. Recruiters work the halls.
  • Latent Space podcast and Discord — Swyx's community; very high job density.
  • AI Tinkerers — global meetup network; SF, NYC, London, Toronto, Berlin chapters are active.
  • MLOps Community — Slack of ~20K practitioners; jobs channel is high-signal.
  • MCP-related Discords — Anthropic's MCP community, plus tool-specific ones.
  • Hugging Face Discord — for ML / open-source-model jobs.
  • NeurIPS, ICML, ICLR, MLSys, CVPR — academic but increasingly industry-recruiter-heavy.

Direct channels​

  • LinkedIn DMs — most senior roles are filled this way. Be a knowable, postable person before you DM cold.
  • X / Twitter DMs — same idea; X is still where most senior AI engineers hang out.
  • Bluesky — growing for the AI-research side specifically.
  • Cold email to hiring managers — works astonishingly well if (a) you have a specific shipped artifact relevant to their team, (b) the email is 4 sentences not 40.
  • Company careers pages for the specific companies you've listed as targets.
  • GitHub — many OSS AI tools hire from their contributor pools; one merged PR can be the entire interview.

Signals you're "AI-engineer ready" for the job market​

You're ready to apply at the next tier when you can:

  • Talk through a recent feature you shipped end-to-end, including the eval story.
  • Hold a 30-minute conversation about a 2026 AI-engineering tradeoff (RAG vs fine-tune, agent vs chain, closed vs open model, prompt caching strategy, hybrid search design).
  • Explain why your last system prompt is shaped the way it is — and what you'd change if you switched from Sonnet to Haiku.
  • Debug a slow LLM call by inspecting trace data (Langfuse, LangSmith, or equivalent).
  • Have opinions on at least one eval methodology, with a war story behind the opinion.
  • Estimate the monthly cost of a feature design before you build it.
  • Identify three failure modes of an agent design and propose evals to catch each.

A focused job-search workflow that actually works in 2026​

  1. Pick 15–30 target companies across two tiers (e.g., 5 frontier labs as stretch, 15 scaleups as primary, 10 non-AI SaaS as backup).
  2. For each, identify the team you'd join (browse their engineering blog, find the team page, read recent talks).
  3. For each, write one targeted note — to a hiring manager, an engineer on the team, or a recruiter — referencing a specific artifact of theirs and one of yours.
  4. Apply formally in parallel to the same 15–30 — but the warm thread is what gets the interview.
  5. Show up in 2 communities (e.g., Latent Space Discord + AI Tinkerers SF chapter) consistently for 3+ months.
  6. Re-evaluate every 6 weeks. If you're not getting first-round interviews, the bottleneck is either the artifact (no shipped portfolio) or the outreach (cold + generic).

The indie path​

If a job isn't the goal: ship in public, build an audience around your AI work, monetize via paid tools or consulting.

  • The 2026 indie AI scene is real and growing. Founders making $200K–$500K+ ARR from a single AI product are no longer rare.
  • Examples of indie / small-team AI products that hit traction: Friend.com, Granola (now bigger), early Mendable, several MCP server projects, niche RAG apps, voice-agent boutiques.
  • The model that works: narrow audience + sharp product + AI as the differentiator (not the headline).
  • Distribution channels that work in 2026: Twitter/X, ProductHunt, niche Reddit / Discord communities, SEO for specific painful queries, partnerships with adjacent tools.
  • The trap: building an "AI tool" with no audience and no narrow problem to solve. The 2024–2025 wave of "ChatGPT-for-X" products mostly failed; the 2026 winners are narrower and more sharply distributed.
Worked example: a focused 2026 job search

Target list (junior AI engineer, NYC, 2026):

Primary (scaleups, NYC presence): Hebbia, Decagon, Harvey, Mercor, Rogo, Crew, Linear (NYC office).

Backup (enterprise SaaS NYC): Box AI, JPMorgan LLM Suite, Two Sigma internal AI, Goldman AI.

Stretch: Anthropic NYC, OpenAI NYC.

Workflow:

  • Week 1: target list, find one engineer on each team via LinkedIn / their engineering blog.
  • Week 2: ship the missing eval suite on your strongest portfolio project. Write a blog post about it.
  • Week 3: send the blog post + a 4-sentence intro to 15 named people. Apply formally in parallel.
  • Week 4–6: take whatever first-rounds come; ask each interviewer for one piece of feedback.
  • Week 6: review feedback, adjust portfolio or pitch, re-apply to the misses with the iteration.

Result: 4–8 first-round interviews, 2–4 on-sites, 1–2 offers. Compare to "I sent 300 LinkedIn applications and got 2 responses." Same time, ~10x yield.

Highlight: referrals are the single biggest junior-search lever

A referred application to an AI-native scaleup converts to a first-round interview at roughly 5–10x the rate of a cold application in 2026. The "I don't know anyone" objection usually means "I haven't shown up in a community long enough." Pick one Discord (Latent Space, AI Tinkerers, MLOps Community), be helpful for two months, then ask. The referral comes from being a known good contributor, not from cold-DMing strangers.

Common mistakes​

Where people commonly trip up
  • Applying to 300 jobs via LinkedIn quick-apply. The 2026 yield on cold spray is brutal — usually under 2%. A 30-application focused search with referrals out-yields a 300-application spray.
  • Only targeting frontier labs. Anthropic and OpenAI get more applications per week than they hire per year. Use them as stretch; build your career at scaleups and big-tech AI orgs.
  • Refusing "AI at non-AI company" roles for prestige reasons. A senior role at JPMorgan's LLM Suite team or Atlassian Rovo gives you ownership, real eval discipline (regulated forces it), and a paycheck that often beats the equivalent scaleup. Don't dismiss them.
  • Treating "remote" as a low-effort backdoor. Remote-first AI scaleups (Modal, Braintrust, Langfuse, Vellum, Replicate, Baseten) often have higher bars than equivalent in-office teams because the candidate pool is global.
  • Cold-DMing without an artifact. "Hi, I'm looking for a job in AI engineering, can we chat?" has a near-zero response rate. "Hi, I shipped X eval framework over your tool Y and wrote it up here — would love your team's take" converts at 20%+ to an interview.
  • Skipping the community step. In 2026, the highest-yield career move under "first AI job" is often "join one community, ship one post a month, answer two questions a week." Six months of that builds the referral surface that powers the job search.
🤔 Quick checkQuick check

→ Next: Chapter 15 Checkpoint.