The pitch that got my attention
Most conversations about AI agents for small business marketing start with capability. This one started with a price. Runable’s co-founder and CEO Umesh Kumar put it to TechCrunch in about the bluntest terms possible: if you are paying an agency $10,000 to run your Google Ads, can someone do the same job for less? That is a procurement question, not a technology question, and it is the right one to ask.
It also tells you where the money has moved. The “build me a website” phase of AI tooling is close to settled. I can prompt a decent static site into existence in an afternoon, push it to a CDN, and wire up DNS before lunch. Nobody is going to win a durable business on that alone, because the marginal cost of generating a landing page keeps falling and the tooling to do it is now bundled into half a dozen platforms. Getting someone to visit that page and buy something is the part that stayed hard, and expensive, and human.
So the interesting bet is on the growth side: campaign setup, SEO, social posting, and now visibility inside chatbot answers. Runable raised $21 million in August 2026 to chase exactly that, as reported by TechCrunch (opens in new tab).
What I want to work out in this post is narrower than “is AI good at marketing.” I want to know three specific things. What does the agent actually cost per outcome, once you count your own hours. Who absorbs the loss when a campaign burns budget and returns nothing. And whether the unit economics behind the discount are stable enough that you can build a marketing budget on them. That last one matters more than the demo, and it is the part most reviews skip.
How a browser-scraping startup became a marketing agent
Runable did not set out to sell marketing. Kumar and co-founder Saksham Sarda started it in 2025 as AI infrastructure, building browser technology for scraping data at scale. Then users started doing something unexpected with the browser agent: asking it for slide decks and websites. The team followed the demand and pivoted to a general-purpose agent.
The Series A came in at $21 million, co-led by Susquehanna Venture Capital and Nexus Venture Partners, with existing backers Together Fund and Array VC also participating. It was all-equity primary funding at a $65 million post-money valuation. A few other markers are worth holding onto, because they frame how much of this is product and how much is still ambition. Roughly 1.7 million registered users. A team of 15 in Bengaluru. Largest markets in the U.S., UK, and Japan, with Japan expected to move up quickly.
Fifteen people is a small number for a product that promises to replace an agency function. That is not a criticism, it is the whole thesis: if the agent does the work, headcount decouples from customer count. Whether it holds at scale is the open question, and it is the same question every agent company is answering right now.
The competitive position is uncomfortable in an obvious way. Runable competes with Anthropic and OpenAI, and with coding platforms like Cursor, Lovable, and Replit, while also buying intelligence from some of those same model providers. Kumar’s argument is that the advantage sits in the assembly work rather than the model: infrastructure, analytics, and distribution stitched together so a small business owner never has to integrate five services. I think that is a defensible position for a while. Integration is genuinely tedious, and tedium is a real moat with a short lifespan. It lasts until the model provider decides the integration layer is worth shipping themselves.
AI agent Google Ads automation moves from demo to product
The product now splits into build and grow. Build is what you would expect: websites, apps, presentations, and content from natural language prompts, with deployment and analytics handled inside the platform. Automated website building and hosting are bundled together deliberately, so the user never opens a control panel. Having spent years inside cPanel and WHM, I understand the appeal. Most site owners do not want to know what a zone file is.
Grow is the newer claim. Running ad campaigns, managing social media, handling SEO, and optimising how a business appears in AI chatbot answers. The stated goal is that an owner asks for a number of customers rather than assembling a website, an analytics property, an ad account, and campaigns separately.
Traction, as reported: zero to a $2 million annualised run rate within three weeks of switching on payments in March, and more than 1 trillion tokens consumed over 90 days, with roughly 60% to 70% of that usage from paying customers. Kumar declined to give current revenue or a paying customer count.
Then there is the number that shapes everything else in this post. Runable is running negative gross margins, partly because it subsidises AI usage for customers. Kumar said so directly. That is not a scandal, it is a strategy, and plenty of infrastructure businesses have done it on purpose. But it means the price you are quoted today is a marketing decision rather than a reflection of cost.
TechCrunch’s own test is the detail I keep returning to. Asked to build and deploy a site for a fictional coffee subscription business, set up analytics, and win the first 100 visitors on a $25 budget, the agent built the site and prepared the campaign, then stopped, because an ad account had to be connected first. Cursor hit a similar wall. Runable says it can currently run ads without a connected customer account for ads on ChatGPT, through partnerships it declined to name, describing them as a “soft wedge.”
What this means in practice for site owners
Strip away the framing and you get a fairly clear division of labour. In my experience with these tools, an agent is genuinely good at account setup, keyword research, drafting thirty ad variants you can cut down to five, pulling reports, and the ongoing account hygiene that agencies bill by the hour: negative keyword lists, search term reviews, flagging disapproved assets, spotting a placement that is eating budget. That work is real, it is repetitive, and it is where a lot of retainer value quietly sits.
What still needs you: budget authority, your offer and pricing, conversion tracking that is actually wired correctly, and the judgement to kill a campaign that is technically improving while commercially failing.
Take a plumbing business spending £1,500 a month. The agent optimises towards a “contact form submitted” event. Cost per conversion drops from £30 to £18, the dashboard looks great, and the owner is delighted for six weeks. Then someone checks the CRM and finds most of those submissions are people outside the service radius asking about jobs the business does not take. The agent did exactly what it was told. The signal was wrong, so efficiency made things worse faster. I have watched human-run accounts fail the same way, but a human usually notices the phone is not ringing.
Two structural things to plan for. The subsidised price is a real discount today and a temporary one, so I would not restructure a marketing budget around it without asking what happens when per-seat or per-token pricing resets. And when hosting, analytics, DNS, and ad accounts all live inside one vendor’s agent, switching cost becomes the product. Keep ownership yourself and grant delegated access.
Can these agents actually beat an agency on price?
On execution cost, right now, usually yes. Two reasons, and only one of them is about AI. Inference is being subsidised, and the labour being displaced is billed at agency rates that include overhead, account management, and margin. On outcome per dollar of ad spend, the honest answer is that it depends entirely on whether your tracking and your offer were sound before the agent started spending.
The savings are most real in a specific shape of account. Monthly budgets under a few thousand. One location or one product. Search campaigns where intent is already clear and the job is bid and keyword maintenance rather than persuasion. Bluntly, accounts an agency would hand to a junior with a checklist. If your retainer buys you a monthly PDF and some negative keywords, an agent will do that for less.
An agency still wins where judgement compounds: multi-step funnels with long consideration cycles, regulated categories where ad copy carries legal exposure, multi-market accounts with currency and language splits, and any account where the creative concept drives more lift than bid management ever will. No agent I have used has generated a genuinely new positioning idea for a client. They generate variations on what already exists.
Compare it honestly by adding up the tool cost, the ad spend, and your own hours at a rate you would actually accept, then dividing by qualified leads. Not impressions. Not clicks. Not conversions as your ad platform defines them, which is often generous. Qualified leads, as judged by whoever answers the phone.
What I expect next in agent economics and AI SEO automation tools
This part is my read rather than reported fact. Runable’s economics depend on a cost curve Kumar described as delivering comparable inference quality at close to 10x less cost. I think broad direction is right and timing is the risk. If that curve flattens for even a couple of years while usage per customer keeps climbing, AI agent gross margins stay underwater and prices have to move. My expectation is that we see usage caps arrive before headline prices rise, because caps are easier to sell than increases, followed by tiered access where the capable models sit behind the top plan. Outcome-based billing will get tried and will be messy, since nobody agrees on what counts as an outcome.
I also expect consolidation pressure from above. The model providers supply the intelligence and keep shipping more of the agent layer, and the integration work that makes a product like this valuable is exactly the sort of thing a platform absorbs.
The gap I do not see anyone closing soon is measurement for chatbot visibility. Everyone is selling optimisation for AI answers. Nobody can tell you reliably how many customers arrived because a model mentioned you, which means claims in that category are largely unfalsifiable for now.
Run a bounded test before you cancel anything
Pick one campaign. Cap the budget at a number you can afford to lose entirely, run the agent against it for 30 days, and leave your existing setup live so you have a control.
Before the first dollar goes out, verify conversion tracking yourself. Fire a test lead, follow it through to your CRM or inbox, and confirm the event that gets recorded is the one you care about. An agent optimising towards a broken signal will spend your money efficiently in the wrong direction.
Keep the Google Ads account, the domain, the DNS, and the analytics property in your own name, and give the tool delegated access rather than ownership.
Then do the boring thing that makes the whole test worth running: open a spreadsheet today and write down your current cost per qualified lead, along with how you calculated it. In 30 days you will either have a number that beats it or one that does not, and either way you will be arguing from evidence instead of impressions.