Runable wants AI agents to do the part startups usually ignore: getting customers

Runable just raised $21 million to push a simple but still underbuilt idea: AI agents shouldn’t stop at generating code, slides, or a website mockup. They should help a small business get customers.

That’s the pitch from the Bengaluru startup, which closed a Series A co-led by Susquehanna Venture Capital and Nexus Venture Partners, with Together Fund and Array VC also participating. The round was all-equity primary and valued Runable at $65 million after investment, according to co-founder and CEO Umesh Kumar.

The money matters. The positioning matters more. Runable is trying to sit between today’s coding assistants and the messier reality of running a small business. That means not just building a site, but handling deployment, analytics, ads, SEO, and eventually the whole loop from “I need a business” to “I need customers.”

Building is cheap. Operating isn’t.

The last two years have been good for tools that make software creation easier. Cursor, Lovable, Replit, OpenAI’s Codex, Anthropic’s Claude Code. Different products, different users, same basic promise: type what you want, get something working fast.

Runable started somewhere else. The founders, Umesh Kumar and Saksham Sarda, first built it as an AI infrastructure company with browser tech for scraping data at scale. Then users kept asking the agent to make slide decks and websites. That happens a lot in AI startups. The product drifts toward the workflow people keep forcing it into.

So Runable pivoted into a general-purpose agent. Since launching payments in March, Kumar says the company went from zero to a $2 million annualized revenue run rate in three weeks. That’s a strong number, though like most AI startup metrics it leaves a lot out. Run-rate revenue can move fast when usage is subsidized and customer concentration is high. Still, it suggests demand.

The bigger bet is that the hard part for small businesses isn’t generating an app or a landing page. It’s everything around it.

A site by itself doesn’t pay the bills. Someone has to connect analytics, set up ad accounts, choose a budget, iterate on copy, handle SEO, maybe manage social posts, maybe make sure the business shows up correctly in AI search and chat results. That’s where most nontechnical founders hit friction. It’s also where agencies and freelancers still make good money.

Runable wants to compress that into one agent.

The product is broader than it first sounds

Runable’s current agent can create websites, apps, and presentations from natural language prompts, while also handling pieces of the infrastructure underneath, including deployment and analytics. The company is extending that into what it calls the “grow” side: ad campaigns, social media, SEO, and optimization for AI chatbot discovery.

It sounds broad because it is. That’s the point.

For a small business owner, the workflow usually isn’t “write code” or “make marketing assets.” It’s “set up a functioning funnel.” That means a website, analytics, ad spend, CRM-ish tracking, email capture, and some way to tell whether any of it is working. Those steps live in different products, which is why so many nontechnical users stall out after the first polished demo.

Runable is trying to hide that mess behind one interface. If it works, the user asks for customers, not toolchain setup.

That’s a better pitch than “AI can build websites.” The market is already crowded there. Targeting small businesses also makes sense. Engineers can tolerate stitching services together. A dentist, tutor, or local retailer usually can’t.

The infrastructure is where agents go to die

TechCrunch tested Runable with a fictional coffee subscription business and a modest goal: build and deploy a site, set up analytics, and get the first 100 visitors with a $25 ad budget.

The agent built the site and prepared an ad campaign, but it stopped when it needed an ad account connected. That limitation matters.

Cursor ran into some of the same walls in a similar test. It could prepare the campaign, but it needed access to a Meta Ads account and payment method, plus a separate hosting service to deploy the site permanently. Runable handled more of the website infrastructure inside its own platform, which helps, but it still couldn’t spend the ad budget without an external account.

That’s the friction point for agentic software in production: permissions, credentials, and third-party dependencies. The model may know what to do. The platform still has to do it safely inside someone else’s systems.

This gets hard fast. Running ads, changing bids, publishing content, or touching analytics all require a chain of trust. If the agent can make a mistake at scale, the mistake has a cost. If it can’t act without human approval at every step, the automation story gets weaker.

Runable says it can run ads without the customer connecting their own ad accounts for campaigns on ChatGPT, through partnerships it wouldn’t name. That may be useful early on, but it also means the company is leaning on opaque distribution deals. Those can help adoption. They can also be fragile.

The economics are still ugly

Kumar said Runable saw more than 1 trillion tokens of usage over the last 90 days, with about 60% to 70% of that coming from paying customers. That’s heavy usage, and it helps explain the margins.

The company currently has negative gross margins because it subsidizes AI usage. That’s not rare in this market, but it’s not a small issue either. Every workflow that chains model calls, browser automation, retrieval, analytics, and generation can get expensive quickly. If customers are paying for outcomes while the startup pays per token, per tool call, and per infrastructure event, the gap can get wide.

Kumar says Runable is working with a mix of models, including some it’s building itself, and expects inference costs to fall. He said the company sees a path to delivering the same quality of inference at almost 10x lower cost.

Maybe. Inference has come down hard over the last couple of years, and model routing plus smaller specialized models can cut costs further. But cheaper models aren’t a business model. They’re an assumption. The best-case version of Runable needs unit economics to improve while customer acquisition and product reliability both hold up. That’s a lot to ask.

There’s also a risk in being too model-agnostic. If your product depends on frontier model companies, they can copy chunks of the workflow. OpenAI, Anthropic, and others are all pushing toward better tool use and more complete app-building flows. If they decide small business growth is worth chasing, startups like Runable have to win on execution, distribution, and workflow depth, not model quality.

Why this matters more than a generic agent pitch

Runable’s closest competitors, according to Kumar, are general-purpose agents like Manus and Genspark. That tracks. They’re all chasing users who want results more than code.

But Runable’s angle is narrower and more practical than the “general-purpose agent” label suggests. It’s trying to own the whole stack for a small business outcome. Not just creation, but promotion. Not just output, but distribution.

That’s a sensible direction because distribution is where most AI tools still look flimsy. Generating a site is easy to demo. Driving traffic, managing spend, and proving conversion is much messier. If Runable can make even part of that reliable, it becomes harder to dismiss as just another prompt-to-webapp wrapper.

It also goes after a real pain point: tool sprawl. A typical owner can end up juggling a site builder, hosting, analytics, search console, ad platform, social scheduler, and maybe an email platform. Each one has its own account setup, billing, permissions, and reporting model. That’s manageable for a team. It’s a headache for one person.

Runable’s best argument is that an agent can sit over all of that and reduce the surface area.

The worst-case scenario is simpler. It becomes a brittle orchestration layer that still depends on external permissions, third-party APIs, and a lot of human intervention. At that point, the “agent” label starts to feel optimistic.

The next test is whether it can close the loop

The startup now says it has about 1.7 million registered users, with the U.S., U.K., and Japan among its biggest markets. That’s a decent spread for a company founded in 2025. Kumar also expects Japan to become one of its top markets soon.

Big user counts are nice. Paying users and repeat workflows matter more here.

Runable’s real test is whether a nontechnical customer can ask for a business outcome, then let the system carry it through without duct tape. If that works for one niche, it can probably work for several. If it keeps stalling when a real account, real budget, or real approval is needed, it’s just another impressive demo with a cleaner UI.

That’s the part worth watching. Not the website builder. The handoff from generated content to generated growth.