The Agentic Bridge: Operationalizing Context via Local Orchestration

The software industry is currently misinterpreting the “CLI Renaissance” as it pertains to the new agentic movement. They view the return to the command line as a niche retreat for developers who need to leverage generative AI to its fullest ability. They are wrong. As I argue in “The Local-First Agent Manifesto,” this is not a retreat; it is a security and control imperative that signals the future of productivity.

Note, you can read “The Local-First Agent Manifesto” if you sign up for the Coda MCP Beta. It is published as a success story where Coda’s new MCP Server is thoroughly explored to build some cool stuff.

For Superhuman to win the agentic marketplace, the company must recognize that the legacy retrofitting of “Packs as Agents” is fundamentally flawed. The winning strategy lies in a powerful synergy: Superhuman Go as the local engine that embraces the general nature of the developer-centric CLI, enriched by the highly deterministic Context provided by Coda.

The Engine: Superhuman Go IS the Extended CLI

My manifesto identifies a critical market reality: developers and power users are flocking to local command-line interfaces (like Claude Code, Antigravity, and Codex) because they offer determinism, privacy, and several other benefits that cloud-hosted SaaS agents cannot match.

However, the usability gap for everyday enterprise workers is not solved by dumbing down this capability. It is solved by wrapping it.

Superhuman Go is not merely a consumer product; it is a vehicle that embraces and extends the developer-centric CLI. It operationalizes the power of local scripting and agentic processing with full access to a worker’s personal computers (and devices) for the mass market. By positioning Superhuman Go as an “Invisible Orchestrator” with deep OS hooks, Superhuman provides a near-100 % deterministic runtime environment. This binary does not hide the power of the CLI; it turns the CLI into an integral engine of vast agentic capability that runs right on the user’s metal, insulating them from complexity while retaining the raw power of local execution.

The Fuel: Coda as the Deterministic Context Provider

The manifesto argues that we must reframe Coda’s role. But let me be clear: this is not a pivot away from Coda as the “All-in-One Doc.”

For enterprises and businesses that have adopted Coda, it remains the premier workspace for human collaboration. It is where strategy is written, projects are tracked, and unstructured ideas become structured reality. That value proposition is immutable.

However, in an agentic world, Coda’s role expands significantly. It becomes the ultimate source of Context.

Agents hallucinate when they lack boundaries. They fail when they lack state. Coda’s unique architecture—mixing text, tables, images, and logic—makes it the “high-value data repository”. It is the only platform capable of feeding the local agent (Superhuman Go) with highly deterministic context. The synergy is simple but profound: users continue to collaborate in Coda, and that collaboration generates the structured “fuel” (invoices, decision logs, wikis) that the Superhuman Go engine consumes to perform accurate work.

The Architecture: Local Orchestration, Cloud Supply

As near as I can determine, the “Packs as Agents” strategy is obsolete not because of where it runs in the cloud, but because it attempts to force agents to reside in the cloud, introducing latency and vulnerability. It also seems particularly vulnerable against an army of modern CLIs that generate code as needed - just in time - guided by dynamic contexts.

I think we’ve proven (with Superhuman Go) that productivity is fundamentally tied to the machine where the work happens—the desktop. The laptop, etc.

Superhuman should adopt the Local-First Orchestration model:

  1. The Brain (Local): Superhuman Go acts as the “Consumer”. It runs locally, reading Markdown skill files that define its capabilities. This eliminates the friction of “heavy, fragile” proprietary Packs.
  2. The Supply (Cloud): Coda acts as the “Utility”. Through the Coda MCP (Model Context Protocol), Superhuman Go queries Coda for live data.
  3. The Dynamic State (Cloud): Coda serves as the active “System of Record” and shared memory. As Superhuman Go executes tasks locally, it writes state changes, decision logs, and process outcomes back to Coda tables in real-time. This ensures that while the execution is local and private, the results remain collaborative, keeping the human team aligned without requiring them to leave their “All-in-One” workspace.

This architecture solves the “Distance Problem”. The agent executes locally with zero latency, manipulating local apps (email, browser, code), but it makes decisions based on the “Truth” stored in Coda.

The Strategic Imperative: Open the Binary

The manifesto’s “Superhuman Opportunity” is the key to market dominance. The existing “Grammarly/Superhuman binary” is a dormant giant with an installation base of over 40 million users - mostly knowledge workers - everyday people who get shit done.

To win, Superhuman must open the architecture to accelerate expansion. Instead of locking users into a closed, high-friction ecosystem of Coda Packs, the platform should allow Superhuman Go to ingest standard, local Markdown skill files—the same standard used by Frontier model providers. The “skills economy” has begun.

By doing so, Superhuman achieves two things:

  1. Universal Compatibility: Every skill built for the open agentic ecosystem becomes usable by Superhuman immediately.
  2. Contextual Supremacy: The platform combines that universal skill set with Coda’s proprietary context.

Conclusion

This is how Superhuman wins.

The company does not win by encouraging more code in a “SaaS Walled Garden”. It wins by building the bridge. Maintain Coda as the unmatched layer for human collaboration and context. Unleash Superhuman Go as the local orchestrator that acts on that context. This is already Superhuman’s positioning. How they achieve it is what I question.

The future is a local engine fueled by cloud truth. As a user deeply invested in this ecosystem, I believe this manifesto is not just an opinion; it is the roadmap to making Superhuman the operating system of the agentic age.

3 Likes

I am with you @Bill_French and I hope that the Coda team can see what you have in mind.

2 Likes

I’m hoping the Coda team comes back and tells me I’m absolutely wrong, because the development of agents in Coda is based on code fragments that are established in your local instance of Superhuman. Even if they’re not installed into it, it would undoubtedly be really cool if they could run within that instance, and I hope that is the case.

However, there is also this fundamental nervous tension about how other agents work and why Coda MCP plays a significant role in bridging the world of agentic systems and tooling with Coda and the work we do in Coda. We cannot ignore that agents are increasingly writing code just-in-time, as needed, for specific purposes. This transient code capability is becoming more pervasive, and it is likely to be a familiar pattern: code that springs up and evaporates as needed.

The entire premise of Packs seems to run counter to that pattern.

This fundamental question is what’s driving me to ask:

Do we need packs at all to achieve our greatest dreams of Coda?

Perhaps Packs play a role, but to me, they appear at the end of their life cycle, given the new world of agentic work we do every day.

2 Likes

This is a compelling argument for local-first orchestration. One parallel that stands out is how power-user tools succeed when they prioritize determinism, local execution, and transparency rather than abstraction-heavy cloud layers.

A good real-world example is Equalizer APO — it operates directly at the system level, stays fully local, and gives users precise, deterministic control without sacrificing performance. That same philosophy maps well to your vision of Superhuman Go as an “invisible orchestrator” that embraces the CLI instead of hiding it.

Wrapping powerful local engines with thoughtful UX, while keeping execution on the user’s machine and context grounded in a trusted system of record, feels like the right long-term direction for agentic systems.

1 Like

Bingo! It does feel like the right future. I hope (and trust) that Superhuman leadership is leaning into the idea that they already have the CLI-centric infrastructure in place in SuperhumanGo.

1 Like