Most companies are stuck on AI chat

While AI use has skyrocketed, many adopters haven’t progressed beyond ChatGPT-style tools, according to a new survey, with some experts suggesting the limited deployment of other AI technologies shows a lack of maturity.

Forty percent of US businesses are getting the bulk of their value from ChatGPT-style tools, while 13% are getting the most value from agents and 10% from custom AI models, according to a survey commissioned by agent platform vendor Decidr.

At the same time, 44% of those surveyed say their organization’s primary use of AI consists of standalone tools used by individual employees, with only 25% saying their AI is integrated into specific processes or workflows, and 18% saying they have a centralized AI platform deployed across the whole business.

Nevertheless, nearly nine in 10 of the 1,200-plus decision-makers surveyed say they expect AI to have a greater impact on their organizations over the next year.

Several AI experts say the survey’s results fit with what they see in the market, with most organizations still stuck in chat AI and standalone tool mode. But they disagree about the impact, with some saying that businesses are missing out on advanced AI uses.

Other AI leaders say standalone and chat tools can bring value and can be steppingstones toward more expansive AI uses.

Search with better prose

The Decidr reports shows that many companies aren’t using serious AI tools, says David Brudenell, co-CEO there.

“Most organizations aren’t using AI; they’re using a very fast search engine that writes back,” he says. “ChatGPT-style tools are retrieval with a polished surface. You ask, it answers.”

Using chat-style tools requires that humans decide what to ask, interpret the response, and route the response to somewhere useful, Brudenell adds. “That’s not automation,” he says. “That’s assisted Googling with better prose.”

Moving to agents allows organizations to achieve new levels of efficiency, he says.

“A GPT answers a question about an invoice,” he adds. “An agent receives the invoice, checks it against the purchase order, flags the discrepancy, routes it to the right approver and logs the exception — without being asked. The difference isn’t speed — it’s who initiates, and where the work stops.”

Instead of employees focusing on what they can ask the AI, they should pivot to asking what the AI can do without them, Brudenell recommends. “The first produces productivity gains at the individual level,” he adds. “The second produces operational leverage at the enterprise level. They compound very differently over five years.”

However, agents aren’t always the obvious next step for some companies, Brudenell says. Most enterprise agent deployments today are probabilistic systems sitting on top of critical processes, he observes.

“That’s genuinely dangerous without proper orchestration and guardrails,” he adds. “The companies that have moved too fast have learned this painfully. Automation that fires incorrectly at scale causes more damage than a slow human process.”

Enthusiasm gap

The survey shows high excitement about AI but also may understate a gap between enthusiasm and integration, says Derek Perry, CTO at AI-native engineering solutions provider Sparq. More than eight in 10 respondents say their organizations understand the power of AI, but only a quarter have integrated it into specific processes or workflows, he notes.

“That’s the most telling data point in the entire report,” he says. “Understanding what AI can do and understanding what it takes to make AI operational are two very different things.”

As Perry works with customers, the AI bottleneck isn’t literacy or ambition, but the condition of underlying systems, Perry says.

“Most organizations are sitting on fragmented data, manual workarounds, and workflows that were never designed to support real-time decision-making,” he adds. “You can’t layer agents or custom models on top of that and expect durable results.”

It therefore doesn’t surprise him that 44% of organizations are primarily relying on standalone tools used by individual employees. “That’s the path of least resistance,” he says. “It requires no integration, no data architecture, and no process redesign.”

These standalone tools are where the return ceiling is lowest, he says, but they’re also a reasonable starting point to drive linear ROI. As such, Perry doesn’t see the comparison of using chat-style AI tools vs. agents as a debate about maturity.

“GPT-style tools aren’t immature — they’re incomplete as an enterprise strategy,” he says. “They’re extraordinarily useful for individual productivity. Summarization, drafting, research, code assistance: These tools deliver real value and I’d never discourage adoption.”

However, chat-style tools have a ceiling, he adds. Standalone tools don’t learn from a company’s operational data, and they don’t enforce business rules, he says.

“They don’t integrate into the decision chains where the actual financial and operational leverage exists,” Perry says. “The maturity spectrum isn’t really about the sophistication of the AI model. It’s about the depth of integration into the work that matters.”

Different tools for different jobs

Philipp Burkhardt, AI team lead at Kingspan Insulated Panels CEME, also doesn’t consider using chat-style tools instead of agents as evidence of IT immaturity.

“They’re different tools for different jobs,” he says. “A carpenter isn’t less mature for using a hammer instead of a CNC machine.”

Standalone tools can give organizations broad, flexible value across the whole organization for low effort, while agents and integrated AI can provide deeper value in specific workflows but cost a lot more to build, maintain, and govern.

Many employees at Kingspan Insulated Panels CEME are using chat-style AI tools to draft emails, summarize documents, and brainstorm, he says. The company is deploying AI chatbots across our websites, building an HR agent for its Czech team, and piloting AI voice agents for handling inbound calls.

“We have standalone tools, process-specific integrations, and early-stage agents all running in parallel,” Burkhardt says. “The standalone tools deliver value today. The agents and integrations are where we think the bigger value is, but they’re harder and slower to get right.”

Chat-style tools can be a good way for a company to experiment, he adds. “The real mistake I see is companies skipping the standalone phase and jumping straight to custom agents before they even understand where AI helps their people,” Burkhardt says. “The more integrated stuff takes significantly longer to deliver and requires way more organizational buy-in.”