Forward deployed engineering
Forward deployed engineering should scale like software.
The least scalable role in enterprise software is growing as AI is sold faster than anybody can deploy it. The repeatable work belongs in a system.
Understand the customer. Design the fit.
The repeatable work. Taken care of.
- Package
A version for each customer
- Authorize
The permissions they approved
- Validate
A real capability, executed
- Operate
Health, changes and recovery
Who is in the middle
Sold as a product.
Delivered as a project.
Something has to sit between the two.
- Software vendor
- Forward deployed engineer
- Enterprise customer
A person, needed again for every customer.
- Software vendor
- Meetext
- Enterprise customer
Software. Judgement calls arrive with their context.
Eleven steps. Every customer.
The job, broken down
- 01Understand requirementsHuman judgment
- 02Understand the environmentHuman judgment
- 03Plan the deploymentHuman judgment
- 04ConfigureMeetext supports
- 05Handle security and admin approvalMeetext prepares · customer approves
- 06DeployMeetext supports
- 07ValidateMeetext supports
- 08DiagnoseMeetext supports
- 09RepairMeetext supports
- 10UpgradeMeetext supports
- 11OperateMeetext supports
AIThe Meetext Agent
The forward deployed engineer, as somebody you can ask
Steps four through eleven are the same motions every time, so the agent does them when you ask. Each change waits for your confirmation.
Meet the agent- “We signed Globex. Add them and send their admin the install link.”
- “Every morning, tell me who is waiting on an admin.”
What is software, and what is not
Software today
- Deployment packaging
- Capabilities frozen into a package per customer.
- Entitlements
- What this customer bought, enforced at the package.
- Authorization lifecycle
- OAuth, admin consent, reconnection, revocation.
- Security review
- Generated, with provenance on every claim.
- Validation
- A real capability executed before anyone is told it works.
- Credential management
- Per environment, rotated before expiry.
- Health
- Three layers, watched separately because they disagree.
- Incident handling
- Fingerprinted, deduplicated, correlated with changes.
- Versioning and rollback
- Per customer, to packages that actually worked.
- Operational history
- Append only, replayable, with attribution.
- Requirement capture
- Unmet requests, recorded where they were asked.
- Fleet-wide demand
- The same request across customers, counted.
- Change authoring
- A pull request with evidence and acceptance criteria attached.
- Entitlement remediation
- A request for something you already ship, granted rather than queued.
- Acceptance drafting
- A brief from what the deployment was asked, for the customer to correct.
- Verified closure
- Nothing is called fixed until a case the customer agreed to passes.
- Permissions per person
- Per-teammate permissions beyond their role, obeyed by their agent too.
- Approval before a change
- Deletions in a customer's workspace always wait for a person, and writes can too.
- Personal data masking
- On every tool result before any client or model reads it, under the customer's policy.
- Spend control
- A monthly budget per customer, checked before every model call, and a stop switch.
- First reading of an incident
- A likely cause and next step from the agent, labelled unverified.
Where people stay involved
- Scoping the engagement
- Meetext arrives with a draft; the conversation with the customer is still a person's.
- Reading the review
- Pull request comments are feedback the loop cannot answer yet.
- Expansion
- The evidence is assembled. Somebody still has to make the call.
- Deeper verification
- Generated Python is parsed and its imports resolved. Every other language gets less, and says so.
The order matters
You cannot automate diagnosis of a deployment you have no record of.
Proposes the fix, citing the record.
Recognises a failure it has seen across customers.
What each customer runs and approved, what changed, what broke, and what fixed it.
Questions
What is a forward deployed engineer?
Why does it not scale?
Is Meetext an AI FDE?
Does Meetext replace our solutions engineers?
You build the product.
We take it from here.
Connect a source, approve what ships and send your customer an install link. 2 customer environments free. No card required.