In the race to bring artificial intelligence to the enterprise, a small but well-funded startup is making a bold claim: The problem hindering AI adoption in complex industries has never been the models themselves.
Contextual AI, a two-and-a-half-year-old company backed by investors including Bezos Expeditions and Bain Capital Ventures, on Monday unveiled Agent Composer, a platform designed to help engineers in aerospace, semiconductor manufacturing and other technically demanding sectors build AI agents that can automate the kind of knowledge-intensive tasks that have long resisted automation.
This announcement comes at a critical moment for enterprise AI. Four years after ChatGPT started the frenzy of corporate AI initiatives, many organizations are stuck in pilot programs while struggling to move experimental projects into full-scale production. Chief financial officers and business unit leaders are growing impatient with internal efforts that have spent millions of dollars but yielded limited returns.
Douwe Kila, chief executive of Contextual AI, believes the industry is focused on the wrong constraints. "At this point the model is almost finished," Keela said in an interview with VentureBeat. "The bottleneck is context – can AI really access your proprietary documents, specifications and institutional knowledge? That’s the problem we solve."
Why enterprise AI fails, and what recovery-enhanced generation was supposed to fix
To understand what contextual AI is striving for, it helps to understand a concept that has become central to modern AI development: retrieval-augmented generation, or RAG.
When big language models like OpenAI, Google, or Anthropic generate responses, they use implicit knowledge during training. But that knowledge has a cutoff date, and may not include the proprietary documents, engineering specifications, and institutional knowledge that form the lifeblood of most enterprises.
RAG systems attempt to solve this by retrieving relevant documents from a company’s own database and feeding them into the model along with the user’s question. The model can then respond to real company data instead of relying solely on its training.
Keela helped pioneer this approach as a research scientist at Facebook AI Research and later as head of research at the influential open-source AI company Hugging Face. He has a Ph.D. Is. from Cambridge and serves as an assistant professor in symbolic systems at Stanford University.
But early RAG systems, Keela admits, were crude.
"The early RAG was very crude – take an off-the-shelf retriever, connect it to a generator, hope for the best," He said. "Errors increased through the pipeline. Hallucinations were common because the generators were not trained to stay on the ground."
When Keela founded Contextual AI in June 2023, he decided to solve these problems systematically. The company developed what it calls "unified reference layer" – A set of tools that sit between a company’s data and its AI models, ensuring that the right information reaches the models in the right format at the right time.
The approach has gained recognition. According to a Google Cloud case study, Contextual AI achieved the highest performance on Google’s FACTS benchmark for grounded, hallucination-resistant results. The company fine-tuned Meta’s open-source Llama model on Google Cloud’s Vertex AI platform, focusing specifically on reducing the tendency for AI systems to invent information.
Inside Agent Composer, the platform that promises to turn complex engineering workflows into minutes of work
Agent Composer extends the existing platform of Contextual AI with orchestration capabilities – the ability to coordinate multiple AI tools across multiple stages to complete complex workflows.
The platform offers three ways to create an AI agent. Users can start with pre-built agents designed for common technical workflows such as root cause analysis or compliance checks. They can describe the workflow in natural language and let the system automatically generate a working agent architecture. Or they can build from scratch using a visual drag-and-drop interface that requires no coding.
The company says what sets Agent Composer apart from competing approaches is its hybrid architecture. Teams can combine strict, deterministic rules for high-risk steps – compliance checks, data validation, approval gates – with dynamic logic for exploratory analytics.
"For highly critical workflows, users can choose fully deterministic steps to control the agent’s behavior and avoid uncertainty," Keela said.
The platform also includes what the company calls "One-click agent customization," Which takes user feedback and automatically adjusts the agent’s performance. Every step of an agent’s reasoning process can be audited, and responses come with sentence-level citations indicating where in the source documents the information originated from.
From eight hours to 20 minutes: What early customers say about the platform’s real-world performance
Contextual AI says early customers have reported significant efficiency gains, although the company acknowledges that these figures come from customer self-reporting rather than independent verification.
"These come directly from customer evaluations, which are estimates of real-world workflows," Keela said. "The numbers are self-reported by our customers as they describe the landscape before and after the adoption of relevant AI."
The claimed results are nevertheless shocking. One advanced manufacturer reduced root-cause analysis from eight hours to 20 minutes by automating sensor data parsing and log correlation. A specialty chemicals company reduced product research from hours to minutes by using agents that searched patent and regulatory databases. A testing tool maker now produces test code in minutes instead of days.
Keith Schaub, vice president of technology and strategy at semiconductor test equipment company Advantest, offered support. "Contextual AI has been an important part of our AI transformation efforts," Schaub said. "The technology has been introduced to multiple teams at Advantest and select end customers, saving meaningful time across tasks ranging from test code generation to customer engineering workflows."
The company’s other customers include semiconductor giant Qualcomm; ShipBob, a tech-enabled logistics provider that claims to achieve problem resolution 60 times faster; and Nvidia, the chip maker whose graphics processors power most AI systems.
The eternal enterprise dilemma: should companies build their own AI systems or buy one off the shelf?
Perhaps the biggest challenge facing contextual AI is not competing products but the trend among engineering organizations to create their own solutions.
"The biggest objection is that ‘we will build it ourselves’" Keela accepted. "Some teams try. It sounds exciting to do, but it’s exceptionally difficult to do well at scale. Many of our customers started with DIY, and find that 12-18 months later they are still debugging recovery pipelines instead of solving real problems."
The company argues that the alternative – off-the-shelf point solutions – presents its own problems. Such tools are rapidly deployed but often prove inflexible and difficult to adapt to specific use cases.
Agent Composer attempts to occupy a middle ground, offering a platform approach that combines pre-built components with extensive customization options. The system supports models from OpenAI, Anthropic, and Google, as well as Contextual AI’s own Grounded Language Model, which was specifically trained to be faithful to the retrieved content.
Pricing starts at $50 per month for self-service use, with custom enterprise pricing for larger deployments.
"The CFO’s rationale is really about increasing productivity and getting them into production faster with their AI initiatives," Keela said. "Every tech team is struggling to hire top engineering talent, so making your existing teams more productive is a big priority in these industries."
The road ahead: multi-agent coordination, write action, and the race to build hybrid AI systems
Looking ahead, Keela outlined three priorities for the coming year: workflow automation with actual write actions in enterprise systems rather than just reading and analyzing; Better coordination between multiple specialized agents working together; and faster expertise through automated learning from production feedback.
"Here the compound effect matters," He said. "Every document you capture, every feedback loop you close, those improvements pile up. Now it will be difficult to catch the companies building this infrastructure."
The enterprise AI market remains extremely competitive, with offerings from major cloud providers, established software vendors, and numerous startups all chasing the same customers. Whether contextual AI’s bet on context over models will succeed depends on whether enterprises share Kiela’s view that foundational model wars matter less than the infrastructure around them.
But there is a certain irony in the company’s situation. For years, the AI industry has focused on building larger, more powerful models – spending billions of dollars in the race to artificial general intelligence. Contextual AI is making a cool argument: for most real-world work, the magic isn’t in the models. It’s in knowing where to look.
<a href