Every idea people wish had an agent.

competitor research

Every PM has a stale Notion doc titled "competitive landscape". An agent that keeps it fresh would be quietly enormous.

contract review

Lawyers were the first profession to be told "ChatGPT will replace you" and the last to actually feel it.

customer support

Every B2B SaaS has a Slack channel for support escalations. Half of them are now "powered by an AI agent".

expense reports

Expense reports are the canonical "why does this still exist?" workflow. Every fintech promises to kill it. Most just OCR the receipt.

hiring

The recruiter-stack arms race: every ATS now ships an AI screener. Most of them screen out the same candidates a human would.

onboarding

Onboarding agents are the moat that every HR-tech founder thinks they have. They mostly do not.

performance reviews

The performance-review cycle: where humans spend two weeks writing the thing an LLM can draft in 90 seconds.

calendar scheduling

Calendar agents are solved. Reclaim, Motion, Clockwise, Cal.ai. If you ship another one, it had better do something they do not.

document search

Enterprise RAG is a $5B category and most of it ships as "the search bar in Notion / Confluence works a bit better now".

email triage

Inbox-zero agents are how Superhuman stayed a $20-a-month product. The bar is high; the second mover usually loses.

meeting summaries

Meeting-summary agents are the saturated frontier. Fireflies, Otter, Gong, Granola, Cap. Pick a niche or do not ship.

slack summaries

Slack already ships an AI summary. Anything you build on top has to clear that built-in baseline.

task prioritization

Every productivity app has tried "AI prioritizes your day for you". Almost none of them survive contact with a real backlog.

accounting

Bookkeeping is the highest-leverage AI agent target nobody wants to talk about because the SMB sales cycle is brutal.

financial forecasting

Every Series-A startup wants an AI CFO. None of them want to pay for one that actually works.

fraud detection

Fraud is the one finance vertical where "AI agent" was already true in 2014. The bar is "beat Sift, Sardine, and Stripe Radar".

invoicing

Freelancer invoicing is the "five-billion-vertical-SaaS" problem. Each vertical thinks theirs is the special one.

personal budgeting

Mint died. Monarch, Copilot, YNAB ate the corpse. AI budgeting agents have to be more than "categorize my Whole Foods charge".

tax prep

TurboTax is a $2B moat made of state-by-state regulation. Your AI tax agent inherits all of it on day one.

code review

Code-review bots are saturated at the surface and wide open underneath. The winners pick a specific failure mode (flakiness, perf, security).

database migrations

Schema migrations are the boundary between "agent that helps" and "agent that wipes prod". Pick which side of that line you are on.

deployment management

Deployment agents are the rarefied air where "AI agent" actually means something — they have to hold real state across systems.

incident triage

The first AI agent to actually de-burden the on-call rotation will be worth obscene amounts of money. None of them do it yet.

log analysis

Datadog and Honeycomb already ship LLM-powered log search. The bar for a startup here is "show me the bug, do not just paraphrase it".

on-call

On-call agents will exist in five years. The question is who gets there first without paging someone in the middle of a real outage.

meal planning

Meal-planning AI agents have died in droves. The good ones own the grocery integration; the rest are recipe blogs in a trench coat.

medication reminders

Medication-reminder agents look trivial until you realise the win condition is "elderly users with no smartphone fluency".

mental health

Mental-health AI is the category with the highest stakes and the loosest regulation. Move carefully — Character.ai has already paid the price.

sleep tracking

Oura and Eight Sleep already won the hardware. The agent layer is the next fight, and most contenders are still UI on top of CSV exports.

symptom checking

Symptom-checker AI is a legal minefield with a 0% chance of being a fun shipping experience. The bar is "do not get sued in 50 states".

workout tracking

Strava + Whoop + Hevy + every smart-watch. The wedge for a new workout agent is the coaching, not the tracking.

course creation

Course-creation agents help instructors ship faster. The model is unit economics: $X per course × not enough courses to matter.

exam prep

Test-prep agents are where you actually find willingness-to-pay. High-stakes exams are the only edtech market that prints money.

homework help

Homework-help is a moral grey zone with brutal unit economics. Every kid has ChatGPT free; convincing them to pay you is the whole game.

language learning

Duolingo, Speak, and Pimsleur have already mainstreamed AI conversation practice. Your wedge is the pedagogy, not the LLM.

study planning

Spaced-repetition apps already encode 90% of what an AI study planner would say. The leftover 10% has to be very, very good.

tutoring

Khanmigo and Duolingo Max set the floor. Any new AI tutor has to clear the "is this just GPT in a school uniform?" test.

home maintenance

Home-maintenance agents would be great. They would also need to talk to your HVAC contractor on the phone. So: 2027 at the earliest.

lease review

Lease-review agents are an obvious win that nobody owns. Mostly because the lawyer-shaped legal hole is wide enough to walk through.

pet care

Pet-care AI is the warmest category in the catalogue and the hardest to monetize past "cute AI thing for Instagram".

plant care

Plant-care agents already exist (Planta, PictureThis) and they make real money. Compete on the watering schedule, not the photo recognition.

recipe suggestions

Recipe agents are sympathetic to demo. Convincing a household to actually use one daily is the whole product.

travel planning

AI travel planners are the canonical "looked great in the demo" category. Convert the demo to a booking and you have a real product.

content discovery

Content-discovery agents are basically asking "can we kill the Twitter timeline?" The answer keeps being "not yet".

game recommendations

Steam, IGDB, Backloggd, and OpenCritic already exist. The agent layer would have to know your backlog, your mood, and your time budget.

movie recommendations

Netflix already has a 10,000-engineer recommendation team. Your AI agent has to be doing something fundamentally different.

music curation

Discover Weekly is the floor. The ceiling is "an agent that understands why I want this specific song right now". Nobody is there yet.

podcast discovery

Podcast discovery is genuinely broken and nobody has fixed it. The category is wide open if you can solve the cold-start problem.

watchlist management

Watchlist apps are the canonical "great-for-power-users, irrelevant-for-everyone-else" trap. Pick a side before you ship.

community moderation

Moderation agents are the hardest social problem in tech, dressed up as a weekend hack. They are not a weekend hack.

dating

Dating-app agents are the category most likely to be both extremely funny and quietly dystopian. Pick a verdict accordingly.

friendship management

Friendship-CRM agents are the most awkward demo you can give. They are also, quietly, what most lonely adults secretly want.

group chat summaries

Group-chat summary agents will exist on every messaging platform within 18 months. The startup window for this is small.

networking

LinkedIn outreach agents are the spam vector du jour. Building one will make you money, but make the inbox worse.

social media management

Social-media agents are saturated above the fold and wide-open below. The win is in the niche-vertical playbook, not the all-in-one.

brand strategy

Brand-strategy AI is what every solo founder wants on day one and is unwilling to pay for on day two.

design feedback

Design-critique agents work in the demo and die in real workflows. Real designers want a peer, not a checklist.

logo creation

Logo-generation agents are competing with $5 Fiverr. They are also competing with "ask DALL-E for free". Pick one.

photo organization

Photo organization is the agent category where Apple and Google already won. Anything you ship rents space in their gaps.

video editing

Video-editing agents are the loudest crowded category in AI. Opus Clip, Descript, CapCut, and every TikTok-adjacent tool ships one.

writing

Writing AI is the original GPT wrapper category. The survivors all pivoted hard from "AI writer" to specific verticals.