01: The problem
Most "AI strategy" comes from people who've never shipped AI. Most AI operations come from people who didn't plan for them.
You hired a consulting firm. They delivered a 60-slide deck. The deck recommended you "build an AI capability." Six months later, no one on your team knows what to do with it.
Or you shipped an AI feature. It worked. Then the costs started climbing. The model started drifting. The eval suite never got built. Your team has no idea if the AI is still doing what it's supposed to.
We do both halves of this, and we do them honestly. The advisory work ends in a recommendation we'd actually build, not a deck we'd hand off. The operations work keeps your AI measurably running, with cost controls, drift detection, and a real human on call when something breaks.
02: Right fit?
Is this the right fit?
Best for
- ✓CTOs and founders evaluating "build vs. buy vs. wait" before committing to AI
- ✓Boards or leadership asking honest questions about whether the AI work is going right
- ✓Teams with budget but no AI roadmap, looking for an opinionated plan before they hire engineers
- ✓Companies with live AI features that need ongoing operations, monitoring, evals, cost control
- ✓Teams who shipped AI months ago and now realize nobody is watching it
Probably not
- —Teams who already know what to build (skip the strategy step and hire us to build it)
- —"AI thought leadership," workshops, or innovation theater without intent to act
- —Companies wanting strategy decks they can't or won't execute against
- —Pure compliance audits requiring certifications we don't hold (HIPAA-certified auditor work, FedRAMP advisory)
03: What you get
What you get when we engage
Two clear halves of the service. Most clients use one or the other; some use both.
01AI Advisory (before the build)
- AI audit, what you have, what's working, what's not
- Build vs. buy vs. wait analysis, opinionated recommendation, not a fence-sit
- Roadmap and prioritization, sequenced, with realistic timelines and dependencies
- Vendor and model selection, Claude vs. GPT vs. open-source, off-the-shelf vs. custom
- Architecture review, would your existing stack support the AI you want to build
- Feasibility memos, for specific use cases your team is debating
- Cost modeling, what AI features will actually cost in production at your scale
02AI Operations (after the build)
- Monitoring across every dimension that matters, accuracy, latency, cost, drift, error rate
- Eval suite maintenance, keep the eval set current as your real-world cases evolve
- Model upgrade testing, when a new model drops, we run your evals before you upgrade
- Cost optimization, smaller models where they're enough, caching where it works, batching where it helps
- Incident response, a human on call when the AI does something it shouldn't
- Quarterly reviews, what's working, what's drifting, what to retire, what to add
03Operations we'll take on
- AI systems we built for you (default, natural continuity)
- AI systems your team built, with a one-time onboarding audit
- AI systems another vendor built, with the same audit, plus a frank conversation about whether we can responsibly take it on
04: How we work
How we work
Strategy and operations don't follow the full Production Path, but they share its discipline. Fixed scope. Fixed deliverables. No hourly invoices. No "let's keep talking" engagements.
01
Everything is documented
Every recommendation has reasoning. Every operations decision has a paper trail. If we change a prompt, you know why. If we upgrade a model, you have the evals that justified it.
05: Stack
The stack we work in
Strategy work uses our judgment and your data. Operations work uses production-grade tooling.
- Audit & analysis
- Live walk-throughs of your codebase, prompts, evals, cost dashboards Vendor benchmarks (Claude, GPT, Gemini, open-source models) Architecture review against your existing infrastructure Cost modeling spreadsheets and dashboards
- Evals
- LangSmith, Langfuse, RAGAS, Braintrust Custom eval suites tied to your real-world cases Regression testing on every model change
- Monitoring & observability
- Langfuse, LangSmith, OpenTelemetry traces Sentry, Datadog for runtime and infrastructure Custom dashboards in Grafana, Metabase, or your existing analytics
- Cost & drift detection
- Per-tenant, per-feature, per-model cost tracking Automated drift alerts on accuracy, latency, and behavior Budget caps, rate limits, and circuit breakers
- Incident response
- On-call rotation (with SLA tiers if needed) Runbooks for common failure modes Postmortems with action items, not just narrative
06: Engagement
How an engagement starts
Tier 01
Strategy Sprint
1–4 weeks. Fixed fee.
You bring the question, "should we build X?", "is our current AI work going right?", "what should our 12-month AI roadmap look like?". We deliver an honest, opinionated answer with reasoning, evidence, and recommended next steps.
Tier 02
Audit Engagement
2–3 weeks. Fixed fee.
Deep audit of an existing AI system, yours, your team's, or one you inherited. Output: a written report on what's working, what's drifting, what's costing you, and what to do about it.
Tier 03
Operations Retainer
Monthly. Fixed fee with optional SLA tiers.
Ongoing monitoring, evals, model upgrades, cost optimization, and incident response. Monthly report. Cancel with 30 days' notice. For clients who go through advisory, hire us to build, and stay on retainer afterward, natural continuity, single accountable team.
07: Why us
Why we'll tell you the truth
No conflicts of interest dressed up as strategy
We're an engineering team, not a consulting firm with a partner program. We don't get kickbacks from model providers, vendors, or platforms. If the right answer is "don't build it," we'll say so.
Recommendations we'd actually ship
Every recommendation in an advisory engagement is something we'd implement ourselves. If we wouldn't build it, we don't recommend it.
Operations with a paper trail
Every change we make, prompt edit, model upgrade, eval update, is logged with reasoning. You can audit our work the same way we audit yours.
No retainer lock-in
Operations retainers cancel with 30 days' notice. If we're not earning the fee, you shouldn't have to argue your way out.
08: FAQ
Frequently asked
How long is a typical strategy engagement?+
Most are 1–4 weeks. A focused feasibility memo on one use case lands in a week. A full AI roadmap with prioritization and cost modeling lands in 3–4. A multi-system audit can take 2–3.
What does the deliverable look like?+
A written document, usually 8–25 pages depending on scope, with reasoning, evidence, recommendations, and clearly-flagged tradeoffs. Plus a working session to walk you through it. We'll happily turn the document into slides if you need to present internally, but the source of truth is the written version.
Can we hire you for ongoing operations without a build?+
Yes. We onboard onto AI systems your team or another vendor built. The first 1–2 weeks of any "ops-only" retainer is an audit, we won't take operational responsibility for a system we don't understand.
What's the SLA?+
Tiered. Standard retainer is best-effort with a defined monthly review cadence. SLA tiers add response-time guarantees (e.g., 4-hour acknowledgment, 24-hour mitigation) and after-hours coverage. SLA tiers cost more, only worth it if your AI is mission-critical.
Which AI vendors and models do you recommend?+
Whichever fits the use case. Claude for complex reasoning. GPT for specific strengths. Open-source (Llama, Mistral) when self-hosting matters. We don't have a "preferred vendor partnership", recommendations are based on your evals, your latency budget, and your cost ceiling.
Will the strategy work end in a sales pitch to build with you?+
Sometimes, when the right answer is "build it, and we're a good fit to do it." But not always. Some of our advisory engagements end with us recommending a build we don't do (a different vendor, an off-the-shelf product, or wait-and-see). That's the point of hiring an honest advisor.
Can you advise on AI compliance, governance, or risk?+
Up to a point. We'll give you a clear-eyed view of operational risks (drift, cost, security, prompt injection, data exposure) and recommend mitigations. We don't do regulated compliance certifications (HIPAA, FedRAMP, SOC 2), for those, you need a certified compliance firm.
Can you work with our in-house AI team?+
Yes, frequently. We're often the second pair of senior eyes on an internal team's work. We document properly, surface tradeoffs, and leave the team better-equipped, not dependent on us.