Ada Mira
Code & Agents · Artificial Intelligence
Fewer searches can cost a coding agent more
Columns
- Fewer searches can cost a coding agent more
- The dependencies behind Hugging Face’s unchanged API
- The layer that can say no to an agent’s call
- Take your old conversations through the SDK upgrade
- Tracing a request through DeepInfer
- Which agent is loading that skill?
- Following an agent request through the stack
- Before adding another step to an agent
- Who owns the agent’s data pipeline?
- The connections behind a quantized model’s explanation map
- The same model with a more local division of work
- Where does a coding agent’s work survive its session?
- llama.cpp changes the route of a local-model request
- Changing agent memory independently of the model
- When the refund task moves into the database
- Giving the rules engine the final count in a compliance agent
- A search agent’s new skill: learning where to stop
- An agent can omit a message without losing it
- Codex makes the environment part of the task
- Holo4 puts cross-interface control in the developer’s hands
- Intern-Decision-2B returns a distribution instead of a paragraph
- GitHub's fuzzing agent writes the harness and leaves each crash for a person to read
- Copilot splits review effort into Lite and Balanced
- MiMo-V2.6 opens the weights and moves the bottleneck to hardware
- ZCode 3.14.0 lists a wiki-upload fix; the test is still the bytes that left
- When the instruction file changes, which layer still holds the lock?
- The agent swapped weights once training sat on the repair surface
- A 4.4-month gap moves scarcity from weights to the tool contract
- Cheese toast, a whistle, and who the agent actually answers to
- Before you check in on the agent less often, ask what the gate measured