{"slug": "build-vs-buy-is-it-still-either-or", "title": "Build vs buy: Is it still either/or?", "summary": "A survey of 300 support leaders by Intercom found that while just over a third currently use their vendor's AI as-is, only 2% plan to keep it that way in 12 months, with 37% intending to build their own interface on top of a vendor's AI (up from 11% today) and 22% planning to build an Agent entirely from scratch (up from 9%). Three in four teams (75%) still plan to run a vendor's AI underneath whatever they build, customizing it or layering their own experience on top, which Intercom frames as evidence the choice is no longer build versus buy but which AI stack achieves a given outcome.", "body_md": "Nowadays it can feel like building bespoke software is only a few prompts away, so it’s no surprise to see more businesses tempted to build their own AI tools.\n\nWe surveyed 300 support leaders to understand how they run AI in support today, and how they plan to run it 12 months from now. We gave them four approaches, from least to most built in-house:\n\n- **Vendor AI, used as-is:** a vendor’s off-the-shelf AI product, loaded with their content and policies. Their customers interact with the vendor’s chat widget or interface.\n- **Vendor AI, customized:** a vendor’s AI product, connected to other systems or extended with custom logic.\n- **Vendor AI, own interface:** the vendor’s AI runs behind the scenes, but their customers interact with their business’s own product or interface.\n- **Fully built in-house:** the team builds its own Agent on foundation models from companies like OpenAI or Anthropic.\n\nJust over a third of respondents currently use their vendor’s AI as-is. But when asked how they intend to run it in the next 12 months, only 2% said they plan to keep it that way. Today, 11% have built their own interface on top of a vendor’s AI, but within the next year, 37% plan to. And 9% have already built an Agent entirely from scratch, with 22% planning to in the year ahead.\n\nBespoke Agents are becoming more common, but three in four teams (75%) still plan to run a vendor’s AI underneath whatever they build, customizing it or building their own experience on top. The choice isn’t really “build” *or* “buy” – that’s a hangover from how software used to get made. Instead teams are asking:\n\n- **What outcome are we optimizing for?**\n- **What AI stack will get us there?**\n\nSome examples:\n\nOnce you know what outcome you’re driving towards, deciding whether to build, and what to build, becomes much clearer.\n\n## What we see teams building today\n\n### 1. Hyper-customization\n\nThe simplest form of building we see is teams shaping an off-the-shelf product well beyond its defaults, bending it to their domain, brand, and workflows.\n\nFor an AI support Agent, that means determining where and how the Agent shows up across the customer journey. Teams define how and when it hands off to a human, the tone it takes for different markets or personas, the policies it enforces, the actions it’s allowed to take, and the knowledge it reasons over. While this happens through configuration rather than code, it’s still a type of building, and often closes the gap that used to push teams towards a from-scratch build.\n\n### 2. Connected systems\n\nOne of the most common ways teams build is by connecting their Agent to the rest of the business. Linking it to CRM, billing, product, inventory, or internal tools gives the Agent context and the ability to take actions on a customer’s behalf.\n\nThe bought Agent (like Fin) provides the intelligence, guardrails, and permissions model, while the internal team builds the integrations. We’ve seen in our research that customer support teams on their second (or third) AI Agent look for ones that are readily able to be integrated into their ecosystem.\n\n### 3. Headless: build the front door, not the engine\n\nSome businesses go further, keeping the vendor’s Agent as the backend but building their own frontend on top of it via API. They may opt to do this if they have existing Agents deployed to other parts of the customer journey or their product is an Agent-first interface and they want to create a single, shared UI for all communication. The result is visual cohesion across every touchpoint, which can benefit overall customer experience.\n\nThis kind of building gives teams exceptional customization on customer-facing aspects while outsourcing the Agent’s “engine” to a vendor. By doing so, teams don’t have to worry about aspects of agentic development that are hard to stay on the cutting edge of like model training, retrieval, reasoning, and consistency across channels.\n\n### 4. Micro-builds around the platform\n\nSome of the most interesting building we see is small. AI has turned support teams into builders, and they’re creating sharp internal tools around the platforms they use to make their work more effective.\n\nThese tools tend to solve three problems: team efficiency, visibility into customer and performance data, and gaps between tools in their existing stack.\n\nSome examples we’ve seen:\n\n- **Knowledge auto-sync:** A system that monitors help center articles, alerts when content needs to be updated, and suggests changes. This is especially useful when new products are released.\n- **Ticket sampling:** A sampler that pulls representative tickets per teammate for regular reviews.\n- **Peer-review apps:** An anonymized, gamified way to review other support reps’ work.\n- **Backlog prioritizers:** A tool that enriches snoozed tickets and suggests next steps.\n- **Conversation auditors:** An auditor that flags conversations where a resolution was assumed but never confirmed.\n\nThese builds are worth celebrating. They’re usually about saving time rather than money, and they show teams putting AI to work on the problems that matter most to them. At Fin, we learn a lot from the customers who push our platform to its edges and many of the tools they build help shape the future of our product.\n\nIf you take this approach, one caution: each tool is just one step in the larger, interconnected loop needed to continuously improve an AI system. For example, an audit that flags content gaps and suggests updates should feed into a testing suite similar to Fin’s [Evals and Releases](https://www.intercom.com/blog/announcing-evals-and-releases). That way, you can have confidence that content changes will result in an improvement to performance. With many disconnected micro-builds, you risk fragmentation, which can hinder efficiency and overall performance.\n\n## Considerations when building your own Agent\n\nFor some businesses, building a fully bespoke Agent **may** be the right call. Based on our research, here are some things to keep in mind if you’re considering building and allocating the necessary resources.\n\n### **1. Budget for the whole iceberg, not the tip** \n\nWhen building their own Agents, we’ve observed customers achieve strong initial performance but soon require a dedicated group of engineers just to hold that performance steady. We’ve also seen teams build a great Agent for chat, but then struggle to extend it to email and voice, and end up stitching together several vendors anyway.\n\nWhen budgeting for an Agent build, remember that getting a working prototype or V1 is only 10% of the job. The other 90% is keeping your Agent performing, everywhere you need it.\n\n### 2. **Security cannot be an afterthought** \n\nWe’ve seen cases of builders inadvertently leaking customer data.\n\nSetting up the right guardrails around AI is hard. It requires maintaining data security, privacy, and safe-handoff behavior. For companies handling sensitive customer or financial data, that’s not a detail to leave for later, so ensure your Chief Security Officer (CSO) is [involved](https://fin.ai/blog/blueprint/how-to-talk-to-your-ciso-about-ai-agent-security) in your Agent’s build.\n\n### **3. Plan for the team you’ll have in two years, not the one you have today** \n\nIf building an Agent for customer service is a priority for your business, make sure it’s a long-term priority. We’ve seen businesses get a successful Agent off the ground and then reallocate engineering talent elsewhere, causing a degradation in Agent performance.\n\n### **4. Weigh the opportunity cost honestly**\n\nThe people who can build a great AI Agent are some of the rarest talent in your business. Point them at your support costs and you get a one-time return. Point them at your customers’ problems and you get compounding returns on your core product.\n\n## So, is it still either/or?\n\nFor most teams, no. The future of customer support software (and most B2B software) will be built *and* bought. If teams want to get the most out of AI, they’ll need to be clear about the outcome they want, buy where a problem is already well solved, and save their best engineering talent for what makes them unique.\n\nThat’s not an easy thing to get right, and there’s a lot more to unpack on all of the considerations we’ve outlined here. Over the coming weeks, R&D leaders from across Fin will dig deeper into each one.\n\nIn the meantime, for a deeper look at the economics, read our CTO Darragh Curran’s “[Build vs buy: the high bar for building](https://ideas.fin.ai/p/build-vs-buy-the-high-bar-for-building).”", "url": "https://wpnews.pro/news/build-vs-buy-is-it-still-either-or", "canonical_source": "https://ideas.fin.ai/p/build-vs-buy-is-it-still-eitheror", "published_at": "2026-10-28 18:23:34+00:00", "updated_at": "2026-10-28 18:45:25.078805+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-agents", "ai-products"], "entities": ["Intercom", "Fin", "OpenAI", "Anthropic"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/build-vs-buy-is-it-still-either-or", "markdown": "https://wpnews.pro/news/build-vs-buy-is-it-still-either-or.md", "text": "https://wpnews.pro/news/build-vs-buy-is-it-still-either-or.txt", "jsonld": "https://wpnews.pro/news/build-vs-buy-is-it-still-either-or.jsonld"}}