The FORGE™ method.
Five gates every engagement passes through.
Most engineering firms sell hours. We sell outcomes — defined as productized engagements with fixed scope, fixed price, fixed window. That’s only possible because every engagement runs through five gates we call FORGE: Frame, Outline, Rebuild, Govern, Engage. Skip a gate, and the productized model breaks. Run all five, and the unit economics work for both sides.
5 gates · from "expensive SaaS" or "PRD draft" to production in 4–14 weeks
The five gates aren’t a sales prop — they’re the operating discipline that makes our pricing model work. Productized engagements look impossible from the outside (How can you commit to a fixed price for a custom build? What if the scope shifts?). They’re possible because each gate eliminates a specific class of cost-overrun risk. Frame eliminates undefined-scope risk. Outline eliminates architecture-debt risk. Rebuild eliminates senior-availability risk via AI-pairing. Govern eliminates handover risk. Engage eliminates lock-in risk.
If a competitor is selling you a build but won’t commit to all five gates, the unit economics are hidden somewhere — usually inside hourly billing or future managed-services lock-in.
Productized pricing isn’t a marketing claim.
It’s a discipline.
Frame the outcome.
Most engineering work fails because nobody defined what "done" looks like before code was written. Hourly billing makes this acceptable — every undefined assumption becomes a billed-hour. Productized engagements force a different discipline: write down what will be true at the end, get it signed, then build to it.
- A 30-minute discovery call → fixed-price SOW within 24 hours.
- Outcome stated as a verifiable end-state (e.g. "Klaviyo-replacement live on AWS SES, all 4 existing flows running, deliverability ≥ 98%, total infra cost ≤ $X/month").
- Every assumption that could move the price is enumerated upfront — gathered in a 5-question pre-call form so nothing is discovered mid-build.
- Scope changes after signing: priced as separate addenda, not absorbed into hourly bloat.
Time-and-materials engagements with vague "we'll figure it out together" framing. Hourly billing without a fixed end state. SOWs that say "scope to be refined as we learn" — that's the language of expensive surprises.
Outline the architecture.
Architecture decisions made early determine 80% of long-term cost. We force three decisions in the first week: who bills you for infrastructure, how telemetry will work, and where your data lives. Get these wrong and you spend the next three years paying wrapper margin to whoever locked you in.
- Direct vendor billing locked in: AWS / Anthropic / Stripe / Twilio / Postmark contracts in your name from day one. We don't mark up infrastructure.
- Telemetry-by-default: every shipped build instruments observability (logs, metrics, traces) on day one — not bolted on later.
- Data residency picked deliberately: GDPR / DPDP / HIPAA / state-specific data flows mapped to actual storage regions before the first commit.
- Migration playbook written before the build starts: how the old stack drains into the new one, parallel-run windows, cutover criteria, rollback path.
Letting "we'll figure out hosting later" persist past Week 1. Architecting on top of a vendor that takes a percentage of your revenue (Shopify, per-MTU SaaS, per-resolution agents). Skipping observability because it's "not in scope."
Rebuild on direct billing.
The build itself runs on a single principle: senior engineers (12+ years) AI-paired for 3–5× throughput, committing daily to your GitHub repo. AI is the multiplier on top of senior judgment — never the primary author. The output is production-grade code in your accounts, not a prototype on our infrastructure.
- Daily commits to your GitHub repo from day one — no "we'll hand it over at the end" surprises.
- AI-paired pair-programming: senior engineer drives, AI tools (Claude, Copilot, Codex) provide the throughput multiplier. Quality stays consistent because two reviewers (human + AI) catch each other.
- Tests written for every workflow before the workflow is shipped — not retrofitted at audit.
- Direct vendor billing in your accounts (AWS / Anthropic / Stripe / Postmark / Twilio) — we don't mark up infrastructure ever.
- Weekly demo + decision log: every architectural decision recorded with the rationale, so future maintainers know why.
"Vibe-coded" output where AI is the primary author and the human just rubber-stamps. Hidden infrastructure markups (where the agency takes 20% of your AWS bill). Code shipped to "our staging server" with handover postponed until contract end.
Govern the handover.
Most consulting engagements end the day the contract ends — with a half-finished handover, missing docs, and a runbook nobody wrote. We treat handover as the gate, not the afterthought. The final week is reserved for it. The 90-day warranty is the commitment that keeps the gate honest.
- 90-day warranty on every shipped build: bug fixes free during this window, no questions asked.
- Runbook delivered with the build: how to deploy, how to roll back, how to debug, how to scale, who to call.
- Architectural decision log: every "why we chose X over Y" written down for the people who maintain the system in 2028.
- ISO 9001:2015 quality-system audit applied to every engagement — not a separate process for "important" clients.
- Source-code escrow available on request (Iron Mountain etc.) for procurement-heavy clients who want extra insurance.
Code shipped without docs. Handover that requires "engaging us for a knowledge transfer at $X/hour." Warranty periods so short they expire before anyone actually tries to maintain the system.
Engage long-term (optional).
Some clients hand the build to their internal team and never need us again. Others want a managed-run partner so the team can focus elsewhere. Either path is valid — we don't force a choice. What we DO force: any long-term engagement keeps the client's direct vendor contracts intact. We never insert ourselves between you and the underlying infrastructure.
- Managed-run pricing $1.2K–$28K/mo depending on tool complexity — quoted as a separate engagement, never a contract obligation.
- Your AWS / Anthropic / Stripe contracts stay in your name, always. We manage the operations layer; you own the underlying spend relationship.
- Quarterly health checks: SOC-2-grade access reviews, cost optimization, dependency updates. Written quarterly memo, not a verbal call.
- Cancel anytime. The engagement ends; your code, your repo, your vendor contracts continue. No exit clauses, no "transition fees," no surprises.
"Managed services" lock-in where you can't cancel without losing the system. Vendor markups buried inside the monthly fee. "Strategic partnership" language that hides the fact that you no longer control the infrastructure.
- Upstream Managed — Optional managed run after a rebuild ships.
- AI Pod — A multi-quarter pod IS sustained engagement — different shape, same principle.
Not every engagement runs every gate at full intensity. Strategy is mostly Frame; AI Pod is mostly sustained Rebuild + Engage; Upstream is the only service that runs all five at full intensity. Here’s the map:
| Service line | F · Frame | O · Outline | R · Rebuild | G · Govern | E · Engage | Notes |
|---|---|---|---|---|---|---|
| Upstream | Core | Core | Core | Core | Optional | Every rebuild runs the full FORGE cycle, with optional managed-run. |
| Build | Core | Core | Core | Core | Rare | Productized bundles; clients usually own post-launch internally. |
| AI Pod | Light | Core | Core | Rolling | Core | Sustained Rebuild + Engage. Frame is shared with the client team. |
| Agents | Core | Core | Core | Core | Common | Same as Upstream but for AI-agent rebuilds — managed-run is more common. |
| Strategy | Core | Memo only | N/A | N/A | N/A | Frame IS the deliverable — written memo. No build phase. |
Intensity legend: Core = full gate · Light/Common = partial · Rolling = continuous (no fixed end) · Memo only = deliverable is the framing · Optional/Rare = available but not default · N/A = does not apply.
A framework is only useful if it tells you what NOT to do. Here’s what FORGE explicitly rules out:
-
RULED OUTTime-and-materials engagements without a fixed end-state.
If “done” isn’t defined upfront, neither is the price. We turn down engagements where the client wants T&M billing — not because we can’t do them, but because the entire FORGE discipline depends on Frame being fixed.
-
RULED OUTInfrastructure markups (taking a percentage of your AWS / Anthropic / Stripe bill).
Every vendor contract stays in your name. We don’t mark up cloud, we don’t take a cut of token costs, we don’t bury margins in your monthly invoice. Direct billing is non-negotiable.
-
RULED OUT“Vibe-coded” output where AI is the primary author.
AI is a multiplier on senior judgment. A 12-year senior engineer drives every line; AI accelerates throughput. The reverse — a junior copying AI output into production — produces brittle systems. We don’t ship those.
-
RULED OUTLock-in via “managed services” or knowledge-transfer fees.
Every engagement ends with you owning the code, the repo, the documentation, and the vendor contracts. If you choose Engage long-term, you can cancel anytime. If you don’t, the build keeps running on your infrastructure with zero ongoing dependency on us.
A representative AmLaw 200 firm engagement (anonymised, full case file available under NDA):
- F · FRAME (W0): 30-min discovery → SOW within 24 hr. Fixed price: $24K. End-state: Harvey AI-equivalent workflow on Anthropic + OpenAI direct, 50 attorney seats, audit-trail logging, matter-aware retrieval.
- O · OUTLINE (W1): Direct Anthropic + OpenAI billing in firm’s name. PHI-flow modeled; data-residency confirmed (US-East). Telemetry: per-attorney usage logs, cost dashboards, escalation rules.
- R · REBUILD (W2–10): 8 weeks. Two senior engineers (12+ yrs avg) + AI-pair throughput. Daily commits to firm’s GitHub. Weekly demos with managing partner.
- G · GOVERN (W11): Runbook delivered. ISO 9001 audit applied. 90-day warranty active. Source-code escrow set up via Iron Mountain.
- E · ENGAGE (post-W11): Optional managed-run accepted at $3.6K/mo + token usage. Firm keeps Anthropic + OpenAI contracts in their name.
Result: $686K Y1 reclaim against published Harvey AI list pricing. Same workflow at <5% of the cost. The IT team was certain there was a catch. There wasn’t.
→ ALL CASE STUDIESWant to see FORGE applied to your situation?
A 30-min call: walk through your stack, identify which gates are the highest-leverage starting point, get a fixed-price SOW within 24 hours.
Book a 30-min call →