Agent-led growth: winning customers through AI agents

What an AI agent checks

An agent researching on behalf of a buyer follows much the same route as a person exploring a product: a pricing page, documentation, a sign-up flow and, sometimes, an API. The difference is that an agent may have less room to work around missing information. If your pricing page only says “contact us for a quote”, it cannot complete a straightforward comparison. Your product may drop off the shortlist, even when a human buyer might have clicked further or called.

Three things determine whether an agent can seriously consider your product.

• Discoverability: your product and documentation appear where a researching agent can find them.
• Evaluability: your documentation is factual and complete enough to explain what your product does, what it costs and how it works without a salesperson.
• Actionability: an authorised agent can help start a trial, run a test or complete a request without a mandatory demo being the only route in.

Every required human handover can introduce another point where an agent’s workflow stalls or the buyer moves on.

Why this also matters for more complex software

The clearest examples currently come from developer tools, where agents often have more freedom to act. For software with longer sales cycles, multiple stakeholders or compliance requirements, the same mechanism is more likely to influence evaluation than the purchase itself. AI research can help buyers form a preference before a salesperson joins the conversation. The salesperson’s role then includes confirming that preference, or changing it if the agent has put another supplier first. People still make the final decisions in larger deals, but the conversation may begin with an advantage or disadvantage established before anyone sat down together.

An extra layer in your go-to-market mix

Agent-led growth complements your existing growth channels: self-service, communities, partners, webinars and human sales. HubSpot illustrates how several channels can work together, with free self-service tools, a large partner programme, community events such as its annual conference, and AI tools for prospecting and support. ClickUp also combines a self-service product with AI-assisted sales and a partner network. Agent-led growth adds another layer to that mix.

The measurement challenge will feel familiar. If you look only at the final step before a deal, such as the salesperson who closes it, you miss what happened earlier. Twilio’s “Ask Your Developer” billboard is a useful example: its contribution cannot be reduced to the revenue directly attributed to the sign. Recognition can still make a later sales conversation easier. Agent-led growth faces a similar attribution problem. If you measure only who closes the deal, you may overlook an agent’s earlier role in building the shortlist.

Look beyond revenue directly attributed to an agent. A prospect who requests a trial and already understands your pricing and capabilities may have used an agent to research them. Ask how they discovered and evaluated your product, rather than assuming the source from their behaviour alone.

What this means for your product and documentation

For most SaaS businesses, this means a small number of practical changes, rather than rebuilding the entire go-to-market approach.

• Make pricing understandable, including larger plans. Indicative prices or a clear range are more useful than “contact us”, even when the final offer is tailored.
• Offer a self-service trial or sandbox without a mandatory demo. Requiring a human meeting as the only route to testing creates friction for both agents and buyers.
• Publish current API and integration documentation, rather than making it available only behind a login or through an account manager.
• Keep your website, documentation and actual product behaviour consistent. An agent can compare these sources, and contradictions weaken the recommendation.

This is the same discipline needed to embed AI in your product architecture instead of adding it as a layer on top: predictable interfaces, current documentation and a technical foundation without hidden exceptions. A slow sign-up flow, unclear error messages or a pricing page without useful information also frustrate human users. These are the moments when trust in a product (article in Dutch) is first tested.