A bit on why I keep building here; for anyone wondering whether Superhuman Coda has a place in an AI-first world.
Most of the AI conversation right now is dominated by a real and reasonable fear: hallucination and non-determinism. Can you trust a system that might invent a fact, or behave differently each time you run it, to actually run your business?
That’s the right question to ask — and I don’t think the answer is to keep AI at arm’s length.
It is to orchestrate AI so it reliably achieves a business goal: not a clever demo, but a process a manager can trust day after day.
That problem has two halves, and Superhuman Coda happens to be unusually good at both:
- Deterministic work belongs in formulas. CFL, buttons, and automations give you reliable, auditable, repeatable execution — the same result every time, with nothing invented — for everything that doesn’t need judgment.
- Judgment belongs in inference. Native AI blocks let you place a focused model call exactly where reasoning is genuinely required — and nowhere else, so the unpredictable part stays small and reviewable.
And the Superhuman Go AI engine is excellent at taking a specification of the business workflow we want to automate and generating the Tables, Formulas, and AI prompts to implement it!
Most of my experiments, a set of docs I called AgentixGo (see here) are really just probes into one question: where is the optimal line between the formula and the inference? Coda lets me move that line freely without ever leaving the document. I don’t know another platform where deterministic logic and AI judgment sit this close together, in a tool a business manager can actually read.
Notion has recently launched their “Custom Agents” feature that goes part of the way along this path. But it falls short, and let’s face it; the Notion formula language, and their database model, are way inferior to Coda, and not quite fit for purpose in serious process automations.
The process I have developed (here) exceeds Notion’s custom agents capability by a large margin. Which is just as well, as many of my bigger clients were tempted to migrate to Notion after their recent launch. Now they are happy to stay with Superhuman Coda (phew!).
But the part I’m most excited about isn’t the runtime: it’s the capture.
Drawing on years as a systems analyst, I’m building a single structured table I call the MATRIX.
A manager describes their business process inside it in plain language. From those cells, Superhuman Go generates the supporting models automatically:
- the attributes of each business object
- the state transitions each object can move through
- the processes that reference each object
If you know Systems Analysis, you’ll recognize those as the data model, the entity life histories, and the process decomposition: the three classic views of any system.
The difference is that here they’re generated, and kept in sync, by the Superhuman Go AI engine, and then handed to Coda for execution as a mix of Formulas and AI Columns.
The manager refines detail in a form they understand; the AI automates from the same source of truth.
That’s the real mission: a methodology where a human manager and an AI system meet in the middle; detailed enough for the machine, legible enough for the person: putting serious ROI within reach of managers who don’t trust “vibe coding” by itself.
I’m not posting this to sell anything or ask for anything. I just wanted to reassure this community that the platform you’re already invested in is, in my view, the best foundation there is for building AI-first business solutions. The ceiling here is a lot higher than people realize.
Today, we are concerned with the cost of tokens, the security of sending data to remote LLMs, and the hallucinations and other errors that LLMs are prone to.
But within a very short time, we will be running very capable LLMs locally within our browsers, phones, watches, robots, vehicles, and wearable devices. And the cost of cognition will shrink towards zero.
So the real problem will become how non-technical managers who understand business processes can orchestrate these AIs to reliably automate those processes at the edge where the work gets done.
That’s my mission for 2027.
respect,
Max