At this year’s re:Invent in Las Vegas, AWS CEO Matt Garman made it clear: the era of simply fine-tuning models or relying solely on RAG (Retrieval-Augmented Generation) is coming to an end. In response to the growing demands of enterprise customers who expect artificial intelligence to deeply understand their specific business, the cloud giant unveiled the Amazon Nova Forge platform.
It’s a move that could change the dynamics in the IT partner channel. Nova Forge introduces the concept of open training models, giving companies access to checkpoints of Amazon Nova models. In practice, this means that organisations are no longer doomed to superficially adapt off-the-shelf solutions, but can train their own highly specialised models, dubbed ‘Amazon Novellas’ by AWS. According to Garman, this addresses a key market need – building AI that not only processes data, but fundamentally understands the customer’s knowledge domain.
In parallel, AWS is expanding its portfolio with the Nova 2 series, targeting every market segment: from cost efficiency to complex reasoning. The family includes the Nova 2 Lite model, designed for everyday production workloads, and the flagship Nova 2 Pro. In this case, AWS does not shy away from aggressive comparisons. Garman claims that in tests of instruction adherence and agent handling, the Pro outperforms market competitors such as the GPT-5.1, Gemini 3 Pro and Claude 4.5. Completing the range are the Nova 2 Sonic, focused on speech handling, and the multimodal Nova 2 Ami, which is the first reasoning model to process text, video and audio simultaneously.
AWS’ strategy goes beyond the models themselves, however. A key piece of the puzzle is the Bedrock Agent Core, a platform for building autonomous AI agents. Garman describes the current growth in this area as a tipping point, comparable to the internet revolution. Combining the ability to train their own ‘Novellas’ with an agent architecture is expected to allow companies to create ecosystems that work exactly like their business processes.
The first implementations are already underway. Sony, using Nova Forge to create its own model, has reported a 100-fold increase in productivity in documentation and compliance processes. For integrators and channel partners, the signal is clear: the focus of innovation is shifting from the implementation of AI itself to its deep, structural personalisation.
