On this page
In 1910, American farms ran on 24 million horses and mules. They pulled the plows, and they ate. About 27% of all the cropland harvested in the US that year went to feeding them. A quarter of the farm existed to fuel the tool that worked the farm.
Tractors freed that land, because they didn't eat hay. They ran on fuel, and as the animals disappeared, demand for petroleum went up. The machine took over the chore. The fuel became the new hard problem.
I've spent eight years building tools that do the work, and I keep relearning that lesson. A tool runs on what you feed it, and feeding it is usually the hard part.
GTM is at that point now. Agents are taking over its chores, and the people who run them are moving up to judgment. Both run on data about people and companies. Getting that data live, with no accounts at risk and at a price that works at agent scale, is the hardest problem in GTM today. It's the problem I built RichAPI to solve. This is how I got there.
Eight years of building the hands
I started on chatbots, with Botpress and Botkit. Then I built TexAu, which automates prospecting work on LinkedIn and the web for sales and growth teams.
TexAu taught me where automation breaks. When AI agents arrived, our customers wanted to run more of them, with more accounts researched, more lists built and more sequences written. The agents were ready. What they ran on wasn't.
Every agent that touched LinkedIn needed a logged-in account behind it. That account needed cookies, a proxy and careful pacing. To run ten times the agents, you needed ten times the accounts, and every extra account was one more thing that could be restricted. The agents couldn't scale, because the accounts they depended on couldn't.
That was the point where TexAu broke for me. The automation logic held up fine. The supply under it ran out.
The chore leaves, and the person moves up
Farms, spreadsheets and ATMs all followed the same pattern. The tool takes over a chore. The people doing the chore move to work that needs judgment. And the tool works only as well as its input.
Farm jobs were 33% of US employment in 1910 and 1.2% in 2000.
Spreadsheets did it to bookkeeping. VisiCalc went on sale in October 1979. Since 1980, the US has lost about 400,000 accounting clerks and gained about 600,000 accountants. The adding-up went to the machine. People moved to the what-if questions: what happens to margin if we raise prices, what if we hire two more people.
ATMs did it to banking. Between 1988 and 2004, the number of tellers per branch fell from 20 to 13, banks opened 43% more urban branches, and teller jobs did not shrink. Each branch needed fewer tellers, so banks could afford more branches.
A spreadsheet full of wrong numbers gives you wrong answers faster. A tractor with bad fuel stops in the field. The better the tool gets, the more the input matters.
What's different about agents
Agents take over a wider set of chores than any of those tools: reading, researching, writing, and deciding what to do next. They are also getting better fast.
The cost of running a model at GPT-3.5's level fell more than 280-fold between November 2022 and October 2024. METR measures how long a task an AI agent can finish on its own. That length doubled about every seven months for six years, and since 2023 it has doubled about every four months. Stanford's AI Index puts organizational use of AI at 88%.
On OSWorld, a benchmark of everyday computer tasks, agents went from 12% success to about 66%. They still fail roughly one attempt in three.
Sales is already feeling it. In 2022, sellers spent 28% of their week selling. By 2026 that had risen to 40%, and 54% of sellers had used agents. Sellers expect agents to cut prospect research time by 34% and email drafting time by 36%.
What agents will take, and what they won't
Agents will take the chores: researching accounts, finding the right people, enriching lists, finding and verifying emails, cleaning the CRM, writing first drafts and timing follow-ups. Most of this work is repetitive, rule-based and easy to check, which makes it agent work.
People will keep the work that needs judgment and carries consequences. That means picking the market, deciding what story to tell it and knowing which message is worth sending at all. It also means building trust, owning the outcome when something goes wrong, and making the call to buy or to sell.
Buyers draw the same line. In a Gartner survey of 632 B2B buyers, 61% said they prefer to buy without a sales rep, and 73% said they actively avoid suppliers who send irrelevant outreach. The same buyers still wanted a seller's input on whether a product fits their company. Agents make outreach almost free to send. When sending is free, judgment about what's worth sending becomes the scarce thing.
Gartner also expects more than 40% of agentic AI projects to be cancelled by the end of 2027. My bet is that many of them will fail on what the model was fed.
The fuel problem
GTM data goes bad on its own, because people move. The median American has been with their current employer for 3.9 years, the shortest since 2002, and 22% of workers have been in their job a year or less. In 2024, 63.2 million US jobs ended, an average of 3.3% of workers every month.
Cached databases are how you end up emailing someone who left the company in March. A person on your team might notice the title looks off. An agent won't. It acts on the stale record with full confidence, a thousand times before lunch.
Here is a pattern I kept seeing at TexAu. It's a composite, not one customer. A team builds a good outbound motion. The agent picks the accounts, finds the people and writes copy that sounds human. Then the replies don't come. Some titles are a job behind. Some emails bounce, and every bounce makes the next email look a little more like spam. The team rewrites the copy, tests new subject lines, changes the offer. The real problem sat upstream, in the list the agent was handed.
The supply can also be cut off. On 24 January 2025, LinkedIn announced legal action against Proxycurl, a data API many teams had built on. In July, Proxycurl shut down, with about $10 million in revenue behind it. The teams that built on it had to find a new data layer for every pipeline that depended on it.
Why now
Everyone is building GTM tools right now: AI SDRs, agent builders, enrichment tables, sequencers and research copilots. Each one needs the same thing underneath, which is current data about people and companies. That data has to arrive without putting anyone's accounts at risk, at a price that holds up at agent scale.
That layer is the hardest problem in GTM today. It's GTM infrastructure for humans and agents, and it's the one I chose to work on. It's what I wished TexAu had underneath it when our customers' agents hit the ceiling.
What RichAPI is
RichAPI is one API key for live GTM data. As of today:
- 65+ endpoints: person and company enrichment, people and lead search, email finding and verification, and phone finding. Also LinkedIn posts, jobs and ads, website and tech-stack data, local business, funding and traffic data, YouTube, AI enrichment and data-cleaning utilities.
- Live data: each record is fetched when you ask for it, from the source. Nothing comes from a copy we made months ago.
- 40+ data providers behind one key. For email and phone lookups, RichAPI checks providers ranked by measured hit rate and returns the first answer that passes its quality checks. No match, no charge. Every lookup comes with a receipt that lists each provider tried and the one that answered.
- Nothing for you to run: no LinkedIn accounts to connect, no cookies, no proxies. If you ever had an account restricted while automating LinkedIn, this is the fix.
- REST or MCP: call it from Clay, Bitscale, TexAu, n8n, Claude or your own agent.
Built for agents and for the people who run them
For agents, RichAPI is a set of tools they can call directly. The MCP server at mcp.richapi.ai handles sign-in with OAuth, so Claude or any MCP client can connect without you pasting keys into prompts. For email and phone lookups, the agent doesn't have to know which provider is good at what. The waterfall makes that choice on the server, and the agent gets one answer back.
For people, RichAPI is something you can audit. When an agent runs a thousand email lookups overnight, the receipts show which providers it tried and which one answered. Endpoint prices are published in credits. There are no seats, so adding another agent doesn't mean buying another licence. Agencies can give each client its own named key and see usage per key.
RichAPI has no prospecting UI, and that is deliberate. Deciding who to sell to is your job. Ours is to make sure the data under that decision is current.
These are the rules I'm building it by:
- Live over cached. The record should be right at the moment you use it.
- Charge for results, and show the receipt. Today that means email and phone lookups bill nothing on a miss, and every enrichment counted on the Pass is a profile returned.
- Nothing of yours at risk. No accounts, cookies or proxies on your side, ever.
- One API for the agent and the person. A person can see what each key ran, and the receipt for every email and phone lookup.
- Say who it's not for. If you want to buy a list and blast it, or you need a contractual SLA today, RichAPI is the wrong tool.
Pricing, and the launch offer
You start with 25 free credits and no card. After that it's pay-as-you-go: you buy credits, and each endpoint's price in credits is public.
For teams running enrichment at volume, the launch offer is the Enrichment Pass: 50,000 enrichments a month for $299. One enrichment is one person or company profile returned, single or bulk. Misses don't count. The same 50,000 on pay-as-you-go costs $750 a month. $299 buys about 20,000 at pay-as-you-go rates, so the Pass pays off once you run more than that each month. Below that, stay on pay-as-you-go.
- Billed yearly at $3,588.
- Unused enrichments roll over for one month.
- Email and phone finders run on credits, outside the Pass.
- Founding 25: the first 25 teams to pay get 60,000 a month at the same $299, locked for as long as they renew.
- Change your mind within 30 days and I refund the $3,588 minus $299 for the month you used. No exit fee.
If you write about GTM on LinkedIn, there's a partner rate on the Pass in exchange for two disclosed posts a month. Email support@richapi.ai for the terms.
Back to the farm
By 2000, farms employed 1.2% of American workers. The horses were gone, the chore belonged to machines, and the hard problem had moved to the fuel.
Agents will do more of the GTM work every year. If METR's curve holds, the tasks they can finish alone will keep doubling. The people running them will type less and decide more. Both will run on whatever data you feed them.
I built RichAPI to make that data live, to charge for results where we can prove them, and to keep your accounts out of it, at any scale. Start with 25 free credits, or email support@richapi.ai if you want one of the Founding 25 spots.
Vikesh