Are Companies Actually Deploying Business Agents Beyond Copilots? A Reddit Thread Asks for Production Truth
Most-Screen7458 · reddit · 2026-10-06
A Reddit poster questions how much real agent autonomy exists in production, arguing that most shipped "agents" are still copilots: answering, summarizing, classifying, routing, and drafting.
Key open questions raised:
- When a deterministic service can perform an action, what does putting an LLM (or an A2A hop) in front of it actually buy, beyond latency, cost, and unpredictability?
- In customer support, LLMs handle intent and routing while policy decisions stay as explicit human-written rules — is there a durable pattern for delegating more?
- Where does autonomy stop for write actions? Are teams letting agents execute bounded changes with audit trails, idempotency, limits, approvals, and rollback?
- What does "self-improving" mean operationally — updating prompts and memory, or changing its own policies and code — and are there production systems with evals that catch regressions?
The author wants operating experience: what the agent owns, what stays ordinary service code, where humans step in, and what has failed.
More from coding & agent
- X's nikitabier: Agent identity standards are the key tech problem of the next 5 years — nikitabier · 2026-10-06
- Amp ships Puck, a built-in assistant that spawns and coordinates your coding agents — willcb · 2026-10-06
- Try Omarchy v0.5.0 ships: big stability, performance and GPU acceleration update for Omarchy on Mac — DanWahlin · 2026-10-06
- Alchemy ships wave of updates: Vinext, Prisma ORM, Fly.io blue/green, transactional callbacks — samgoodwin89 · 2026-10-06
- Alchemy Effect 4.0 goes stable: broad GCP coverage, and Neon becomes more than a database — samgoodwin89 · 2026-10-06
- Dev builds a Hermes Agent that runs fully on-device inside an Android phone — alexcovo_eth · 2026-10-06