Will software copilots & agents become an advertising channel for developer tools & APIs, ala Google Search?
A trend I see more and more as I speak with developers that use copilot tools (e.g. Github Copilot, Cursor, Augment) is that they increasingly discover and choose third party libraries, services, and APIs based on their software copilot.
E.g. rather than Google “Video APIs”, find the top few options, read through their documentation, and then implement one, a developer may simply ask their copilot to help them implement a video rendering API. If you ask this of Cursor today, it will list Mux as a “(Recommended)” provider over Cloudflare, AWS MediaStream, or other options.
Cursor recommends Mux as a video rendering service provider
In essence, copilot products are becoming a marketing & distribution channels for developer services.
As these tools mature from mostly assisting with autocomplete and chat-with-your-codebase into agentic systems that provision infrastructure or implement larger scale projects on their own (ala what you see with Replit agent or Cognition), the degree to which this is true will go up dramatically. You will expect the AI to evaluate tool/service options, pick top contenders, and make recommendations on which services to use in order to implement the feature or PRD that they are assigned.
If this plays out, it is inevitable that you will see a formalization of marketing tools focused on the software copilot & agent channel. Companies which offer developer services will be desperate to figure out:
SEO - What is the equivalent of “SEO” aimed at software copilots & agents? What types of content are LLMs good at picking up and understanding? How do you make sure that your service is always listed if someone asks a copilot about your product category?
Ranking - What do you need to do to get copilot products to rank your service as “better” or “recommended” vs. others (as we see in the Mux example above)? How do you ensure that the software agent is factually correct in terms of understanding your offering and comparing it to alternatives?
Analytics -How do I measure what leads or signups are coming from software copilots & agents, and then optimize this metric over time or test changes against it? This is essentially Google Search Console or Ahrefs, but for copilot marketing.
As this space matures, you will almost certainly see a proper advertising market formalize - essentially, what is the Adwords equivalent for software libraries/APIs which want to promote themselves in copilot products? A very simplistic way this could work is as follows:
Advertiser (e.g. Stripe) negotiates deal with copilot vendor (e.g. Github Copilot) for queries related to certain terms (e.g. “Payments API”)
Github Copilot, when a user enters a query or question related to a “Payments API” in the chat window, stuffs marketing collateral relates to Stripe’s payment API into the context of the data sent to the model. It instructs the model to mark any suggested output related to Stripe as “promoted” or to wrap that part of the output in special tokens
The Github Copilot UI visually delineates “promoted” suggestions from “standard” suggestions that the model implicitly came up with from its pre-training data set
Github tracks autocompletions or similar “acceptance” metrics that come from promoted suggestions, and uses that to measure the success of the advertising campaign, analogous to how Meta/Google price on clicks
There would then be a marketplace for bidding on certain keywords across different AI code generation vendors.
There are numerous other ways this could be handled or implemented, and the above solution would have some issues - but I share it just to give a sense of one way it could work even without any major changes at the model layer.
Some of the open questions on my mind in relation to all of this:
Will account creation & payments end up being bundled into copilot products? E.g. if Cursor tells me I should use Mux, why can’t I just sign up for a Mux API key in my IDE via Cursor? If an autonomous agent implementing a PRD wants to use Mux or Neon or Vercel or any other services, does it need to ask me to go create an account and add payment details in all those cases and then give it the API key, or is there a better solution?
Is this any different from non-developer-facing products who want to figure out how to promote their content in ChatGPT? In some sense, the same problem I am describing exists for any SaaS product that wants to be referenced by ChatGPT when someone references a certain term. Why can’t one vendor solve SEO/Analytics/Advertising for LLMs broadly? I do think there are some specific nuances in the developer-oriented version of this, but it is a fair question
LLM Vendors vs. copilot/codegen providers - If an advertising market emerges here, does the supply side directly interact with the demand side (copilot/codegen tools), or do the companies training LLMs need to be directly involved? Today, a lot of service provider suggestions in tools like Cursor come implicitly from the model pre-training - e.g. I doubt Cursor has anything to do with listing Mux as a recommended Vendor. But, over time, I think that advertising in LLMs will look more like in-context learning because you want to be able to dynamically inject or adjust promoted content separate from model training. In this world, the model provider would not be as relevant.
I suspect there are startup opportunities in and around this area - just as you saw with the rise of mobile (AppLovin) and web search (DoubleClick, AdMob) as discovery channels.
