In a world where AI coding agents are growing more autonomous, one startup is making a bold bet: the terminal isn't dying—it's evolving. Amp, the AI coding company that spun out of Sourcegraph in late 2025, recently unveiled a completely rebuilt CLI called Neo. Designed for remote control, plugin extensibility, and long-running agent workflows, Neo represents a philosophical shift in how developers interact with AI. While many argue that agents are moving beyond the command line, Amp insists the terminal remains a vital control surface—just one that needs to adapt. This article explores the seven most important things you need to know about Amp's new CLI and its vision for an agentic future.
1. The Terminal as a Control Surface, Not a Relic
Amp's CEO Quinn Slack and the team argue that despite the rise of autonomous agents, the terminal still matters. However, its role is shifting from a primary workspace to a strategic control surface. Developers can start an agent thread locally in the terminal, then monitor and guide it from anywhere—whether through a web interface or another tool. The terminal becomes one of many touchpoints, not the only one. This approach acknowledges that while agents can operate independently, developers still need a direct, low-latency channel to intervene, inspect output, or issue high-level commands. In Amp's view, the terminal's death is greatly exaggerated—but its form must change to fit an agentic future where developers are orchestrators, not just typists.
2. Remote Control: Manage Agent Sessions from Anywhere
The headline feature of Neo is its remote control capability. When a developer starts an Amp CLI thread locally, they can connect to that same session remotely via Amp's web interface. The system streams live terminal updates into the browser, allowing users to send follow-up prompts, queue messages, interrupt tasks, or cancel the agent entirely—all without being physically at their terminal. This is especially valuable for long-running coding tasks that span hours or days, such as refactoring large codebases or running complex build processes. By decoupling the agent session from the local machine, Amp enables developers to step away, collaborate with teammates, or even travel without losing visibility or control. The remote control brings a new level of flexibility to coding workflows, aligning with the industry trend toward agentic systems that require less constant supervision.
3. A Ground-Up Rebuild for Performance and Scalability
Neo isn't just a cosmetic upgrade—it's a complete architectural overhaul. According to Amp, the original CLI was too tightly coupled to the local terminal session, making remote interaction inefficient. The new design moves the agent loop itself into the cloud, drastically reducing the amount of data shuttled between the server and the client. Slack highlighted on X that this change cuts data transmission by approximately 95%, enabling him to ship features even over airplane Wi-Fi. This efficiency gain is critical for agentic workflows that involve continuous streaming of logs, diffs, and AI reasoning. By running the agent loop server-side, Neo can maintain context across interruptions, handle larger tasks, and respond faster to user inputs—all while keeping the terminal lightweight.
4. Plugin-Powered Extensibility for Custom Tooling
To prevent the CLI from becoming a walled garden, Neo introduces a plugin system. Developers can extend the CLI with custom tools, integrations, and automations—everything from connecting to proprietary databases to adding Slack notifications for build success or failure. The plugin architecture is designed to be modular, making it easy to share and reuse components across teams. Amp envisions a marketplace of plugins that allow the terminal to adapt to any development environment or workflow. This openness is a strategic move to compete with more monolithic AI coding tools, positioning Neo as a flexible hub rather than a fixed interface. Whether you're a solo developer or part of a large enterprise, plugins let you tailor the CLI to your specific needs without waiting for Amp to build every feature.
5. Compaction-First Architecture for Long-Running Agents
One of the biggest challenges with autonomous agents is maintaining context over long periods. As an agent processes hundreds of steps, its internal state can balloon, leading to slowdowns and memory issues. Neo addresses this with a “compaction-first” architecture that periodically compresses and summarizes the agent's history, discarding irrelevant details while preserving essential context. This allows agents to run for hours or even days without degrading performance. The approach is inspired by database compaction algorithms, adapted for conversational and coding workflows. For developers using Amp to automate large-scale refactoring or continuous code reviews, this ensures that the agent remains responsive and coherent, regardless of how long it's been working. The compaction-first design is a technical enabler for the kind of long-running, autonomous agents that Amp believes define the future.
6. The “Coding Agent Is Dead” Paradox
Amp has made the provocative claim that “the coding agent is dead.” What they mean is that the current generation of AI coding agents—tightly bound to a single editor, terminal, and user session—is being superseded by a new paradigm. In the future, agents will operate across multiple environments, be more autonomous, and require minimal direct supervision. This seems at odds with rebuilding a CLI, but Amp resolves the paradox by reframing the terminal as a control surface rather than the agent's cage. The terminal becomes a place to launch, monitor, and debug agents that can roam across GitHub, cloud IDEs, CI/CD pipelines, and more. The “death” of the old agent model clears the way for agents that are both more powerful and more flexible—exactly what Neo is designed to support.
7. Industry Trends: Not Alone in Reinventing the CLI
Amp isn't the only company rethinking terminal-based interaction with AI. Both GitHub Copilot CLI and Claude Code have introduced similar remote control features, allowing developers to monitor and interact with long-running sessions from outside the terminal. This convergence suggests a broader industry recognition that the command line's future lies in becoming a gateway to distributed, cloud-powered agents. By investing in remote management and plugin ecosystems, these tools are competing to become the central hub for developer-agent interaction. However, Amp's approach stands out for its emphasis on architectural efficiency (95% less data) and its controversial yet forward-looking claim about the death of traditional coding agents. As the landscape evolves, developers will have more choices in how they delegate tasks to AI, and the terminal—rebuilt and reimagined—will remain an essential part of the conversation.
With Neo, Amp is betting that the terminal can evolve into a powerful, remote-controllable, and extensible hub for autonomous coding agents. By decoupling the agent loop, embracing plugins, and designing for long-running tasks, they aim to give developers the best of both worlds: the immediacy of the command line and the flexibility of cloud-native tools. As AI coding agents become more autonomous, the terminal's role will shift—but it won't disappear. Amp's vision is clear: the terminal still matters, but only if it's rebuilt for an agentic future.