Foundation model
A foundation model is a large pre-trained AI model designed to be the base layer for many downstream applications. Frontier LLMs (Claude, GPT-4, Gemini) are foundation models. The same weights power summarization, code, reasoning, and conversation.
Why it matters
The 2010s were the era of task-specific neural networks. Each task (translation, sentiment, classification) had its own model, trained from scratch on a curated dataset. The 2020s replaced that with foundation models: one big pre-trained model that handles many tasks via prompting and fine-tuning.
For AI products built in 2025-2026, picking the foundation model is the most consequential design choice that does not affect the product's UX directly. Different foundation models have different strengths:
- Claude (Anthropic): strong on long-context reasoning, instruction-following, and tool-use API.
- GPT-4 (OpenAI): strong on general conversation, code interpreter sandbox, image generation.
- Gemini (Google): strong on multimodal (image + audio + video) and Google-tool integration.
Most modern agents pick one as the default and route to others when the task calls for it. The foundation model is the brain; everything around it (memory, tools, channels, identity) is the product layer.
The competitive landscape moves fast. New model releases happen every few weeks; capability differences shift quickly. Production agent products typically abstract the foundation model behind a routing layer so they can swap or mix without changing user-visible behavior.
How Ajeris handles this
Ajeris uses Claude as the default foundation model. The choice is deliberate: Claude's tool-use API and instruction-following behavior are well-suited to the multi-step agent loop. We route to other foundation models when the task calls for it (e.g., image generation, certain coding tasks).
The user does not see the foundation model directly. They see "Ajeris answered." The model is invisible until something visibly changes (a capability lift from a model upgrade, a new channel becoming possible). We absorb the upgrades silently.
