How FDE Deployment Works

Deploying an FDE is not installing a tool. Once a 0ai engineer completes the initial setup, the FDE begins operating autonomously. The tenant PM’s role is to review the FDE’s observations and approve the scope of automation.

Deployment, Not Installation

Just as Palantir’s FDE embeds within organizations to solve problems, Deprex’s AI FDE is “forward deployed” into client organizations.

The deployment flow:

1. A 0ai engineer executes the initial deployment — tenant environment creation, Slack connection, daemon startup
2. The FDE begins autonomous observation — monitoring Slack conversation patterns, learning workflows
3. The PM reviews observations — receiving task patterns and proposals via Slack
4. Automation is approved incrementally — Shadow Comparison validates quality, PM approves production displacement

PMs do not need to run commands or edit configuration files.

What Happens During Initial Deployment

The 0ai engineer handles the following. PMs just review the results.

Tenant Environment Creation

app/orgs/acme/
  skills/          # Where the FDE stores auto-generated skills
  rules/           # Tenant-specific behavioral rules
  memory/          # Autonomously accumulated business knowledge
  config/          # Authentication and environment settings
  scheduled-tasks.yaml  # Scheduled task definitions

This directory structure is created by the 0ai engineer. Once the FDE starts operating, memory/ and skills/ are automatically populated.

Slack Connection and Daemon Startup

When the FDE’s Slack daemon starts, the following begins automatically:

• Receiving @mentions in designated channels
• Observing conversation patterns in the channel
• Auto-executing scheduled tasks

@fde Summarize only the pricing-related questions from this week's inquiries

PMs simply talk to the FDE in their regular Slack channels.

Automatic Initial Memory Construction

After deployment, the FDE automatically learns organizational context from:

• Past messages in Slack channels
• Shared documents and spreadsheets
• Meeting transcripts (when tools like Tldv are integrated)
• Business knowledge gained from conversations with the PM

PMs do not need to manually write initial memory files. The FDE auto-generates them from observations and asks the PM for confirmation.

The PM’s Role: Observation and Approval

While the FDE operates autonomously, the PM does three things:

1. Review Observations

The FDE reports detected task patterns via Slack.

[Observation Report] Detected a pattern: support backlog is manually
compiled every Monday. Shall I generate a skill to automate this?

The PM simply responds with “yes,” “no,” or “observe a bit longer.”

2. Approve Automation Scope

When the FDE auto-generates a skill, it presents Shadow Comparison results to the PM.

• Human output and FDE output are compared side by side
• Production displacement is proposed when quality is equal or better
• No transition to automated execution without PM approval

3. Receive Results

Results from automated tasks are notified via Slack. Anomalies or items requiring attention are highlighted separately.

The Autonomous Cycle After Deployment

Once initial deployment is complete, the FDE continuously runs the following cycle:

Observe --> Learn --> Skillify --> Displace --> Improve

• Observe: Detect task patterns from Slack conversations, meetings, and workflows
• Learn: Record patterns as task flows and accumulate them in memory
• Skillify: Auto-generate skills (SKILL.md + scripts) from task flows
• Displace: Production displacement after Shadow Comparison validation and PM approval
• Improve: Auto-collect execution feedback and self-improve skills

PMs receive notifications at each stage and provide approval or corrections as needed.

Automatic Scheduled Task Proposals

The FDE also detects patterns for recurring tasks.

[Proposal] Add a scheduled task for weekly KPI summary on Fridays?
Recommended schedule: Every Friday at 16:00
Notification channel: #acme-fde

When approved, the FDE auto-adds it to scheduled-tasks.yaml and begins execution the following week.

FAQ

What if the FDE proposes incorrect automation? Just say “no” in Slack. The FDE remembers the feedback and won’t repeat the same proposal.

How long until results appear? Slack-based request processing works immediately after deployment. Autonomous task displacement proposals typically begin after 1-2 weeks of observation.

Does the PM need technical knowledge? No. Slack conversations and approving/rejecting notifications are the PM’s primary interactions. Configuration file editing is handled by the 0ai engineer.