Companies evaluating customer-facing AI agents tend to focus on what they can do. How reliably does the AI understand the request and retrieve the right information? Can it complete a transaction without handing the customer to a person?

But AI deployments often fail for reasons that don’t have anything to do with the capability of the software. Instead, they stall because the company is lacking the expertise to successfully set it up. How do you configure the prompts, access the knowledge sources, set up the integrations, and build the appropriate guardrails and evaluation tools required for customer service operations?

Unfortunately for many AI systems, this set-up work still requires specialized expertise. Large companies can hire consultants or forward-deployed engineers. But smaller companies often struggle through a pilot and never make it into production.

An Agent To Build and Operate the Agents

Typewise, which provides an AI agent platform for customer experience, has seen this expertise gap in most of the mid-sized companies it services. To fill that gap, the company has launched Nova, which it describes as an AI operator for its customer experience platform. Nova isn’t an AI agent that interacts directly with customers. Instead, its job is to build and manage the collection of Typewise agents that do.

A company connects its systems and describes the desired customer experience in ordinary language. Nova can then create specialists for areas such as billing, returns or warranty claims, determine which systems they need to access and draft the instructions (prompts) that govern their behavior.

The job continues after deployment. Nova monitors production activity, investigates changes in performance and identifies possible causes. If customer satisfaction falls following a policy or workflow change, for example, Nova can diagnose the problem, propose a revision and simulate the change before asking the company to approve it.

Typewise Isn’t Alone

NiCE Cognigy recently introduced a related approach called Agentic Building for CX AI. Its open-source Agent Plugin allows coding agents including Claude Code, ChatGPT, Codex and Gemini to work directly with a Cognigy environment.

The plugin gives those agents access to current Cognigy documentation, platform tools exposed through MCP, Cognigy-specific skills and specialized agents for tasks such as building an AI agent or auditing a voice configuration. A user can supply a transcript, requirements document, URL or statement of work and ask the coding agent to create the corresponding flows, knowledge stores, tools and endpoints. The same agent can test the implementation and prepare it for deployment.

The two approaches currently differ in presentation and intended user. Cognigy’s capability is primarily an implementation partner for developers, architects and solution consultants working through an external coding assistant. Typewise positions Nova as a built-in operator that a customer service team can address directly and that remains involved throughout the production lifecycle.

Software Starts Absorbing Services, But Does the Platform Survive?

Looking beyond the world of CCaaS, agentic solutions like the recently released Grok Bot take an even more radical “no code” approach. Once a user installs the Grok Bot platform, she can simply instruct her first agent, who she might call “Chief of Operations,” to build itself and all the agent personas it thinks are required to carry out the work she wants done. The decades-long promise of no code solutions has finally arrived.

But the advent of self-configuring AI agents isn’t just about closing the implementation gap.

A more radical possibility is that increasingly capable software agents reduce the need for standardized commercial applications altogether. This was the thinking behind earlier “SaaSpocalypse” warnings, which admittedly haven’t really panned out.

The issue with bespoke enterprise software was that it was expensive to build and maintain, complex to deploy, and wasn’t necessarily as good as an off-the-shelf best-in-breed application. It made no sense for a company to cobble together their own ERP or CRM systems when experts had already fine-tuned business processes based on best practices and encoded it all in relatively affordable (everything is relative!) applications.

Could software agents change that calculation? Instead of buying a complex CCaaS platform that contains many more features than it needs, a company might describe the customer experience it wants to AI agents. Then it could connect its existing systems to the agentic platform, provide access to its policies, interaction histories and operating constraints. Agents could design and build the required specialists, workflows, integrations, evaluation systems and management tools specifically for that organization.

Suddenly the company would have a bespoke CX application. So, have we arrived at a time and place where such one-off systems make sense?

Probably not, at least not in the foreseeable future. Who, though, can see beyond a few days in our current era of exponential technology acceleration, even if calls to pace the frontier slow progress?

But having software that can reliably and expertly configure itself? That seems like one of those capabilities that might soon become the new table stacks. Implementation expertise is becoming a product feature.

Categories: Articles