methodology

How a Forge build
actually runs.

Two-week or six-week — the cadence is the same. Discovery up front, async build in the middle, structured handoff at the end. No surprises, no Tuesday standups, no scope creep.

stage 1 — discovery

Day 1-5: workflow audit + signed Statement of Work.

Discovery Week is the most important week of the engagement. Almost every failure mode in AI agent implementation traces back to under-specified scope. We refuse to start build until Exhibit A is signed by both sides.

What we do

Recorded screen-share of the current workflow (your team performs it; we observe). Test data review. Edge case enumeration. Success criteria definition. Integration access scoping. Exhibit A drafted Wednesday, signed Friday.

What you commit

One single point of contact. ~3 hours of workflow shadowing. Sample data set (50-200 records with labelled correct outputs). Integration credentials (read-only first). Approval of Exhibit A by Friday EOD.

stage 2 — build

Day 6-21: agent ships into your stack, async.

Build phase is where most agencies hide; Forge is the opposite — every weekday gets a Slack update, every Friday gets a working demo. You see the artefact on real data, not slides.

Week 2

Agent core + first integration

Agent built into your stack against scoped credentials. End-of-week demo on the real workflow. Edge cases logged for Week 3 polish.

Week 3

Remaining integrations + sandbox tuning

Tuned to ≥80% success rate on the agreed test set. End-of-week demo on production-shaped data. Customer's QA team gets sandbox access for parallel testing.

stage 3 — ship

Day 22-28: production deploy + structured handoff.

Code transfer, runbook, monitoring dashboard, training session. The agent doesn't leave our laptops as a black box — your team gets enough to maintain it.

Code yours

Repo transfer on day 28

Private GitHub repo transferred to your org. MIT-licensed code, full git history, all dependencies pinned, CI configured.

Runbook

Self-service maintenance guide

15-page runbook covering: prompt edits, integration credential rotation, model upgrades, error patterns, when to escalate. Written for a mid-level engineer to maintain.

Training

Live walkthrough + Q&A

4-hour training session for your team (recorded). Covers architecture, common changes, dashboard interpretation. Q&A continues async via private Slack for 30 days.

stage 4 — ongoing

Day 29 onward: maintain it yourself, or upgrade to Care.

After the support window closes, you have three honest paths. We tell you upfront which one usually fits.

Self-maintain (default)

Your team owns the code + runbook. Prompt tweaks, integration credential rotations, model upgrades — all documented and explainable. Most teams fit this path comfortably.

Cost:$0/mo. Just your engineer's time (typically <1 day/quarter).

RECOMMENDED

Care plan (Q2 2026 launch)

Monthly health check, prompt drift monitoring, model upgrade migrations, 5 hrs/mo of agent tweaks included. We watch the metrics; you focus on the product.

Indicative price: $200-400/mo (final pricing announced at launch). Joining the wait-list now reserves Q2 introductory rate.

Phase 2 build

When the agent's working and you want to extend scope (more integrations, additional use cases, custom UI), we treat that as a fresh engagement under the same flat-fee model.

Cost: Sized per Phase 2 SoW; typically $15-30k for incremental work.

non-negotiables

Things we won't do.

See the methodology applied to your workflow.

Send a brief — the use case in plain English. We'll respond within four hours with a Statement of Work draft.

Send a brief →Back to home