Learn · Articles
AI Stack S5: Email as the front door
In May I said I was testing email as the front door to our AI workflows, so other people on the team could benefit from the same system without a terminal, a new app or installing anything. I said I would report back. Here it is.
01 · The layer
Layer 5 is the only layer built for everyone
This is layer 5 of the stack I sketched out in June — the only layer built for people who never touch the tools.
The idea is simple: the same files, process guides and tools I use with AI agents should also be available to colleagues through ordinary email.
02 · The interface
One address, an allowlist and a scope per sender
One internal address is the interface. An allowlist decides who may use it. Each sender has their own scope: which files the agent may read, what changes it may propose and a daily request cap. For analysed requests, a separate model audits the drafted reply against that scope before the runtime sends it.
The request arrives as a normal email. The agent reads our actual working files, follows the documented process, does the work and replies in the same thread with any attachments.
03 · The results
91 requests from 21 April through July
From 21 April through July, it handled 91 requests. The work included country-level biowaste and BSF analyses, feed arithmetic, filtered internal contact spreadsheets and briefing documents in Word. It also replied in the sender's own language, which mattered more than I expected.
When a request exposed a larger knowledge gap, the assistant created a defined research task instead of improvising an answer.
04 · The lesson
The benefit and the rollout lesson
That is the practical benefit of this layer: colleagues do not need to learn the agent software, our file structure or how to write a long prompt. They ask a normal business question. The system carries the context, process and boundaries.
The rollout lesson was equally practical. People use it when it solves a recurring task faster than asking me—not because I announced a new AI interface.
05 · The point
Bounded access turns a founder's system into a company capability
My thesis through this series has been that your automation level is set by your documentation, not your tooling. Layer 5 adds something important: bounded access turns a founder's system into a company capability.
If I introduced it again, I would start with three recurring tasks for each person and show the first useful result.
If your team uses internal AI agents, what made them part of normal work?
Keep exploring
Related topics and pages
- Articles library — the written guides this post joins
- BSF business course — where documented assumptions shape a case
- BSF start-up course — the validation steps a small team can run with shared records
Read next
- One fleet, different local realities — the standard-versus-configurable split behind the control layer
- AI Stack S4: Agent software and connectors — the layer below this one, on agent software and connectors
- AI workflows for the whole team (email as interface) — the earlier experiment with email as the team interface
- Our AI stack, demystified — the layered overview this series extends
Next step
Put the team behind the front door
Manna can help organise the files, guides and tools behind a BSF case so the same workflow reaches the whole team.