How an SES firm uses an AI development & operations team
There is a lot of talk about filling a developer shortage with AI, but leaving out "what it can and cannot take on" erodes trust. Here it is split into can, conditional and cannot.
FDE Box is not a tool to replace an on-site person — it is delivery capacity for development, infrastructure and operations. An AI multi-agent team executes the steps; humans approve anything hard to reverse. So knowing exactly "what it can do" is where use begins.
Selling "AI replaces on-site staff"
Explaining it that way to customers or the field loses trust on deals with presence or command requirements.
No boundary on what to hand over
Without splitting can / conditional / cannot, you deploy on the wrong deal and fail.
Security and liability feel uncertain
It is unclear whether customer source or personal data enters external AI, or who owns incident liability.
Confusing "AI tool" with "execution team"
FDE Box is not a chat tool but an execution system for development and operations with steps and an approval flow.
Governance explained only abstractly
Unless "enterprise governance" is made concrete as data location, no-training use, separation and audit logs, there is no basis to judge.
The human approval points are invisible
Without showing which work a human approves, automation looks risky.
Apply the opinion’s §3 distinction directly. Making the boundary explicit is what builds trust.
What it can do
Requirements shaping, design support, coding, test automation, Dev/Stg/Prod setup, CI/CD, monitoring, incident analysis, performance tuning, cost optimization, documentation.
Conditional areas
Attending regular meetings, direct access to customer systems, handling personal data, 24/7 incident response, internal customer approvals — applied after agreeing conditions.
Areas hard to replace
Named-person on-site presence, direct command of an individual, a legally named responsible party, physical on-site work, internal organizational coordination.
Enterprise-governance checklist
- Data storage location and per-customer data separation
- Whether data is used for external model training (no-use by principle)
- Access control and audit logs
- The list of actions a human must approve
- Whether source code may leave, and how data is deleted at contract end
Frequently asked questions
Is AI deployed into an SES deal like a person?
No. It does not replace parts needing on-site presence or direct command. It takes the executable deliverable areas of development, build and operations, and applies conditional areas after agreement.
Do you feed customer source or personal data into external AI?
Data location, no use for external training, per-customer separation, access control and audit logs are defined under enterprise governance, and the scope is agreed per deal.
Who is liable for incidents or judgment errors?
Humans approve anything hard to reverse. Liability scope is set in advance by contract type (quasi-delegation / contract / operations).
Validate the AI dev & ops team with a real deal
Send one deal sheet where feasibility is unclear. We judge which areas we can take and send it back.