Twenty-five jobs ranked by AI-agent exposure between 2026 and 2030, with methodology, the human skills that survive each, and the ten net-new roles AI is creating. This is the honest version — not the doom-monger version, not the techno-optimist version. The transition is real, slower than the hype and faster than the skeptics expect.
This article is for people making career decisions, leaders making team-design decisions, and policymakers thinking about labor in the AI agent era. It's based on the 88 agents we've reviewed, our methodology and a careful read of which agent capabilities are actually shipping in production today versus living in vendor decks.
For the broader context see agentic AI vs generative AI, autonomous vs copilot agents, when not to use an AI agent and our AI agents that work overnight coverage.
How "exposed" is measured
We rank each role on five axes:
- Task automatability — what fraction of the role's daily tasks an agent can do well today.
- Quality threshold — how close to human-level the agent needs to be before substitution makes economic sense.
- Cost-of-error — what happens if the agent is wrong. High cost-of-error slows replacement.
- Regulatory friction — laws that require a human in the loop.
- Relationship gravity — how much the role depends on trust and rapport.
A role with high task automatability, low cost-of-error, low regulatory friction and low relationship gravity is high-exposure. A role with the opposite mix is low-exposure even if some tasks are clearly automatable.
The 25 most-exposed roles, 2026–2030
| Rank | Role | Exposure | Why | Survivable angle |
|---|---|---|---|---|
| 1 | Outbound SDR | Very high | Cold-email + research + reply handling all solved | Strategic, top-100-accounts sales |
| 2 | Tier-1 customer support | Very high | Most tickets are deterministic policy lookups | Tier-2/3 escalation, voice of customer |
| 3 | Routine bookkeeping | Very high | Categorization + reconciliation = solved | Forensic, advisory, controller work |
| 4 | Manual data entry | Very high | OCR + extraction is mature | Data engineering, quality control |
| 5 | Junior paralegal research | High | Doc review + summarization is strong | Discovery strategy, witness prep |
| 6 | Content moderator | High | Triage of obvious cases automated | Edge-case judgment, policy work |
| 7 | Scheduling assistant | High | Calendar + inbox solved | Executive Chief of Staff work |
| 8 | Mid-volume copywriter | High | Most marketing copy is good-enough | Brand voice, long-form, strategy |
| 9 | Junior coder (autocomplete-scope) | High | Cursor, Claude Code accelerate | Senior architecture, code review |
| 10 | Routine financial analyst | High | Spreadsheets, basic models automated | Story-telling with data, M&A |
| 11 | Translator (general) | High | High-quality machine translation in 2026 | Literary, legal-binding, simultaneous |
| 12 | Travel agent (basic) | High | Itinerary planning agents work | Luxury, complex, expert curation |
| 13 | Insurance underwriting (low-complexity) | Medium-high | Standard policies automated | Complex risk, fraud, broker work |
| 14 | Medical transcription | Medium-high | Solved | Clinical documentation strategy |
| 15 | Tax-prep (simple returns) | Medium-high | TurboTax-class automation extends | Complex returns, advisory |
| 16 | Mortgage processor | Medium-high | Doc-heavy, deterministic-rules-heavy | Underwriting judgment, problem files |
| 17 | Recruiter (high-volume sourcing) | Medium | Sourcing + outreach automated | Top-of-funnel relationship work |
| 18 | Procurement analyst | Medium | Catalog match, simple negotiations | Strategic sourcing, supplier rels |
| 19 | Market research analyst | Medium | Synthesis automated | Hypothesis design, executive presentation |
| 20 | Editor (entry-level) | Medium | Copy-edit + style is automated | Acquisition, narrative shaping |
| 21 | Inside sales (transactional) | Medium | Outbound + qualification automated | Complex deals, channel partnerships |
| 22 | Compliance analyst (routine) | Medium | Policy lookup, audit prep automated | Investigations, regulatory strategy |
| 23 | Quality control auditor (digital) | Medium | Pattern matching at scale | Root-cause analysis, process design |
| 24 | Tutoring (basic) | Medium | One-on-one explanation is workable | Coaching, motivation, special-ed |
| 25 | News reporter (commodity) | Medium | Earnings summaries, sports recaps automated | Investigative, narrative reporting |
The "exposure" column is task exposure, not job-elimination probability. Most of these jobs change shape rather than disappear by 2030.
The 10 jobs AI agents will likely create
| New role | What they do |
|---|---|
| AI agent engineer / architect | Build the agents themselves |
| Prompt + eval engineer | Maintain prompts, run evals, prevent regressions |
| AI ops / agent SRE | Run agents in production, monitor observability |
| AI safety + red-team specialist | Test agents against prompt injection and abuse |
| AI product manager (agentic features) | Decide what agents in products should do |
| RAG + retrieval engineer | Build the knowledge stack under agents |
| AI policy + compliance officer | Run governance and compliance for agents |
| AI-augmented domain specialist | Lawyer/doctor/accountant who runs at 3–5× productivity |
| AI training data engineer | Curate datasets for fine-tunes and evals |
| Human-in-the-loop reviewer | Approve agent outputs in high-risk loops |
A useful observation: the new roles favor people who combine domain depth + agent fluency. The single safest career trajectory in 2026 is to deepen your domain expertise and become the best in your company at directing agents to do that domain's work.
Roles that are mostly safe through 2030
Five categories where AI agents are co-pilots rather than replacements:
- Skilled trades. Electricians, plumbers, HVAC, master carpenters, mechanics, machinists. AI helps with diagnostics, scheduling and parts, but the hands-on work is unautomated.
- High-trust professional services. Strategic sales for top accounts, senior management consulting on novel transformations, M&A advisory, executive search. The relationship is the product.
- Care and mental health. Therapy, social work, geriatric care, hospice. AI assists with intake and documentation but the human bond is the work.
- Senior leadership. CEO, founder, head of department roles where the work is judgment over scarce information, hiring, and culture-shaping.
- Creative IP authoring at scale. Long-form journalism with deep sourcing, fiction with distinctive voice, novel research at the field's edge. The output is "this person's perspective"; an agent can't replace the person.
The honest transition view
Three things to internalize, regardless of which list your role is on:
Partial automation, not full replacement. Through 2030, the dominant outcome is 70–90% of routine tasks in a role being automated, with the human role compressing toward judgment, relationship and strategy. That's still a major change but not the same as "the job disappears."
Within-role inequality grows. The top quartile of any role — the people best able to direct and audit AI agents — get more productive and more valuable. The bottom quartile compresses with the automation. The middle bifurcates.
Sector-by-sector unevenness. Customer support, SDR and bookkeeping will be deeply transformed by 2030. Healthcare and education will move slower because of regulation and human-trust gravity. Public sector slowest of all.
What individuals should actually do
Three durable moves, in order of leverage:
- Become the agent's user. Whatever you do, learn to direct AI agents to do most of it. Domain depth + agent fluency = outsized productivity. See how to use AI for research, how to use AI for SEO content, how to use AI for cold outreach and similar guides.
- Move upstream. Judgment, strategy, relationship and novel problem-solving are slower to automate. The compressing parts of every role are repetitive and rule-bound; the expanding parts are upstream.
- Specialize in evaluation and governance. Domain experts who can audit, red-team and approve AI output are scarce in 2026 and will be scarce in 2030. Becoming that person in your domain is unusually durable.
The generic "learn to code" advice has aged poorly because junior coding is itself in the exposed column. The advice that has aged well: deepen one domain, become fluent in directing agents within it, and develop the judgment to spot when the agent is wrong.
What organizations should do
Four shifts visible in the best-run AI-agent adopters in 2026:
- Don't fire and replace; redesign. The teams winning don't lay off; they redesign work and redeploy people upstream.
- Invest in eval and observability before scaling. Without these, you can't tell if the agent is doing the work well.
- Pay for upskilling explicitly. A meaningful share of headcount should rotate through agent-engineering or human-in-loop reviewer roles for 6–12 months.
- Be honest with employees. Vague promises about "AI will only augment" age badly. Specific plans (which tasks, which timeline, which transition support) age well.
See our methodology for how these organizational signals affect our scoring of the agents themselves, and how to pick an AI agent for procurement guidance.
What policymakers should think about
This article is operational rather than political, but three observations worth noting:
- Income transitions during occupational reshuffling are the meaningful policy variable, not job counts in aggregate.
- Education systems need shorter feedback loops between curriculum and labor demand. The 4-year-degree model lags badly here.
- The compliance frameworks we already have (EU AI Act, GDPR, NIST AI RMF) work for now; the gap is enforcement capacity, not law.
The bottom line
AI agents in 2026 are real, useful, and will substantially change ~25 occupations over the next 5 years. They will not eliminate work, and they will create roles that didn't exist in 2024. The people who do best in this transition aren't the ones who fight the tide and aren't the ones who let agents replace them passively — they're the ones who become the agent's most effective user in a domain they already understand deeply.
The hype is louder than the reality, but the reality is still big. Plan for it.