■ HIGH RISK ■ Sports & Entertainment
The data half of the job — which horse, which race, which trainer is live — is exactly what software does better. What survives is the hustle: relationships with trainers, managing a rider's career and ego, and being trusted at 6 a.m. on the backside. AI won't replace jockey agents so much as shrink what they can charge for.
“AI matches jockeys to races by data. Your Rolodex is a museum piece.”
Our AI replacement risk score — how we score jobs
A jockey agent's product is the 'book' — hustling mounts for one or two riders by working the condition book, calling trainers, watching morning works, and playing a daily chess game: committing a rider to a horse in Thursday's fifth means turning down another trainer whose goodwill you'll need next month. The agent lives on the backside at dawn, trades in gossip and favors, and takes a cut of the rider's earnings, so every open mount is rent.
The analytical layer is being commoditized. Past-performance data, pace figures, and trainer-pattern statistics that agents once carried in their heads are now queryable by anyone; entry platforms have digitized the condition book; and it's short work for modern tools to rank which available mounts give a rider the best win probability. A trainer with a data platform needs an agent's opinion less than one who handicapped by feel. That erodes the information edge that let sharp agents outperform — the same handicapping brain that justified the fee is increasingly a subscription product.
But racing runs on relationships and trust in a way few industries still do. Mounts are promised on a handshake, and a trainer gives a live horse to an agent who's been straight with him for a decade — not to whoever's algorithm ranks highest. Agents also do career management software can't: talking a rider through a slump, deciding when to leave a circuit, smoothing over a bad ride with an angry barn. The job's deeper risk isn't AI at all: it's racing's own contraction, with fewer tracks, shorter meets, and fewer mounts to book. Our 55 score is mostly about a shrinking pond, with software draining the analytical moat.
Automatability: our editorial assessment of current and near-term AI capability
Serious pressure by around 2030, from two directions at once: data platforms erasing the agent's analytical edge, and racing's ongoing contraction shrinking the number of mounts to book anywhere. The relationship core of the job automates last — trainers will still give horses to people they trust — but fewer tracks and shorter meets mean fewer agents overall, each needing a bigger book to make the same living.
It's replacing the handicapping half — software already matches riders to live mounts better than memory and hunch. But mounts in racing are secured through trust and long relationships with trainers, which no platform provides. The realistic outcome is fewer agents, each leaning on relationships and service while the analytical edge becomes a commodity anyone can subscribe to.
At major circuits with a strong rider, yes. The tighter constraint isn't technology, it's the sport: fewer tracks, shorter meets, and smaller foal crops mean fewer mounts to book each year. Agents who survive tend to handle top riders, work multiple circuits, or offer real career management beyond hustling entries.
Get the mount. Data can tell you the four horse in Saturday's feature is live; it can't convince the trainer to put your rider on it over the rider he owes a favor. Add slump psychology, circuit strategy, and repairing relationships after a bad ride, and most of the job's value turns out to be diplomacy.
Use them as a weapon, not a threat. Agents who combine trainer trust with sharp data — knowing which conditions suit their rider, which barns are hot before it's obvious — book better mounts than either the old-school phone hustler or the pure numbers approach. The tools are cheap; the relationships took a career. Keep both.