Skip to main content
Orbit’s ambition spans several technically demanding fields. A serious plan acknowledges the constraints rather than treating them as footnotes.

​
21.1 Model quality

Models can generate incorrect information, misinterpret instructions, or fail on tasks that look similar to ones they have solved before. Product design should support verification and make uncertainty visible where it matters.

​
21.2 Tool misuse

An agent can misunderstand a request or misuse a connector. Narrow permissions, confirmation gates, idempotency controls, and reviewable logs reduce risk but do not eliminate it.

​
21.3 Security exposure

Every new connector, API endpoint, and agent capability creates additional security work. Feature expansion should not outrun the team’s ability to test and maintain those controls.

​
21.4 Compute and cost

Training and serving models require compute, power, storage, and operational expertise. Model ambitions must be aligned with available resources. Efficiency work and carefully scoped experiments can help, but cannot erase fundamental resource requirements.

​
21.5 Data quality and rights

Training data may contain errors, sensitive information, or material subject to usage restrictions. Dataset governance, licensing review, filtering, and documentation are necessary components of model development.

​
21.6 Robotics risk

Physical systems can injure people or damage property if perception, planning, control, or hardware fails. Simulation, independent safety controls, supervised testing, and restricted deployment are essential. A language model’s confidence is not a safety certification. Requirements depend on jurisdiction, product function, customer type, data handling, and deployment context. Legal obligations should be assessed with qualified professionals. This document is not legal advice and does not establish compliance.

​
21.8 Organizational capacity

A small team cannot operate every proposed product line at once. The company must sequence work, choose narrow initial use cases, and seek appropriate expertise before entering high-risk domains.