Why birth care stays with people
Midwifery runs on touch, timing and trust. A midwife feels an abdomen to check the baby’s position, reads a laboring person’s breathing and posture, supports the body through delivery, and decides in minutes whether to keep going or call for help. Those are physical, high-stakes judgments made in a room with two patients at once. Software can flag a pattern in a monitoring trace. It cannot put hands on a shoulder dystocia or steady a frightened first-time parent through transition.
The second reason is responsibility. Consent, risk conversations and the choice to escalate to an obstetrician all sit with a licensed clinician whose name goes on the record. A tool can suggest. Someone still has to answer for the call, and for the harm if it was wrong.
What is moving is the paperwork around the birth. Visit notes, discharge summaries, education handouts, appointment reminders and coding are all text tasks, and text is where today’s systems are strongest. That is task erosion, not a job disappearing. The question “will AI replace midwives” is really a question about which slices of the week change hands first.
What AI does, what it helps with, and what people keep
The work this page classes as automatable is clerical and repetitive: drafting routine visit documentation, pulling history into a structured summary, generating standard patient education material, and handling scheduling and reminder messages. AI handles 0% of task time in that group. None of it touches the delivery room.
A larger band is assistive. Reviewing fetal monitoring data for patterns, screening histories for risk factors, checking medication information, and prepping a draft care plan for a midwife to accept, edit or reject. AI helps with 41% of task time. In each case the output is a suggestion a clinician signs off on, which is why the time saved is real but modest.
Everything physical and relational stays with a person: attending labor and birth, hands-on examination, newborn assessment, breastfeeding support, and the long conversations about pain relief, birth plans and loss. Work that still needs a human accounts for 59% of task time here. Coverage, our answer to whether AI can do the job today, sits at 21 out of 100, where a higher number means more of the week’s task time is within reach of current systems. The coverage method explains how that is built.
What the evidence shows, and what it doesn’t
There is no published head-to-head test of an AI system against qualified midwives on midwifery work. Our quality parity grade for this job is D, which is the grade we use when the comparison has not been measured. Because of that, we give no parity number here, and you should treat any claim that a model “matches midwives” as unsupported.
What would settle it is narrow and testable: a prospective study comparing AI-assisted interpretation of intrapartum monitoring against experienced midwives and obstetricians on the same cases, with outcomes reported; a trial of AI-drafted clinical documentation measuring error and correction rates in maternity notes; and a controlled look at triage decisions in antenatal care. Until something like that is published and replicated, the honest answer is that the decision-making core of this job is untested against people. How we grade evidence is set out in the quality parity method.
The labor-market picture is steadier. The Bureau of Labor Statistics counts about 35,010 workers in this occupation, with median pay of $65,790, and projects employment growth of 5.1% over 2025 to 2035 (BLS, 2025). That is a growing job in a system with staffing gaps, not one being planned out of existence.
When this could change
Most likely after 2042 (8 in 10 of our scenarios). Read that as a scenario range rather than a forecast of one date; the replacement year method explains what the window measures and how it is produced.
Two things could pull it earlier. First, tooling cost: the cost figures on this page put AI assistance far below the annual cost of the human hours it touches, so clinics have a reason to adopt the clerical layer quickly. Second, documentation and triage adoption spreading from hospital systems into community and birth-center practice, which would shrink the admin share of the week faster than expected.
Two things hold it back. Roughly half the task time in this job is physical, and the robotics tier it would need is dexterous humanoid work: machines able to handle soft, unpredictable bodies under time pressure. Nothing at that tier is deployed in maternity care. Regulation and liability are the second brake. Consent, scope of practice and malpractice exposure all assume a named clinician, and changing that takes years of rulemaking, not a software release.
What to do: get fluent with the documentation and monitoring tools your employer buys, so you are the person who checks their output rather than the person they are measured against.
How midwives stay needed
Lean into the parts of the week no system is near. Attending labor and birth, including hands-on management of complications. Physical assessment of parent and newborn. The counseling work: birth planning, feeding support, perinatal mental health and bereavement care, where being trusted is the whole intervention.
Two skills raise your floor. One is clinical escalation judgment, the ability to decide fast when a case leaves normal range and to brief an obstetric team clearly. The other is practical oversight of AI output: knowing where a drafted note or a flagged trace is likely to be wrong, and documenting your own reasoning when you overrule it. That skill is becoming part of the job description rather than an extra.
If you are weighing adjacent paths, the closest work sits with nurse midwives, nurse practitioners and obstetricians and gynecologists. You can put any two side by side on the compare page, or see how the wider field is scored in healthcare and across healthcare practitioner occupations. For broader context, the list of jobs that mostly need a person shows what these scores look like across the economy, and the full scoring method shows how each figure on this page is built.