Guide · 8 min read
How to Pick an AI Automation Partner (Honest Buyer Guide)
The takeaway
The signal you want from an AI automation partner: production systems they've actually shipped (not decks), productized scope + pricing, ownership of the deliverable transfers to you. The red flags: blank-check consulting, portfolios that are 80% strategy decks, projects that lock you into ongoing dependency.
Why the AI automation buyer market is broken
In 2024-2025, every consulting firm in the world rebranded themselves as an "AI automation agency." Most of them are repackaged web development shops, marketing agencies, or strategy consultancies with no actual AI engineering depth.
The result is a buyer market where:
- 80% of "AI automation agencies" can't point to production AI systems they've shipped
- Pricing is opaque (almost all agencies refuse to publish rates because their model is custom-scope)
- Deliverables are often slide decks + half-built prototypes
- Buyer due diligence is hard because the category is too new for reliable third-party reviews
This guide is the framework we'd use if we were the buyer.
The five signals that matter
1. Shipped production work. Ask: "Show me 3 AI systems you've shipped that are running in production today." Not pilots. Not prototypes. Not decks. Working systems with real users. If they can't show you 3, they haven't shipped enough to be a credible partner.
2. Productized scope + pricing. Real expertise produces productized offers. If every engagement requires a custom proposal and a 4-week scoping cycle, the partner is making it up as they go. Productized offers ($4,500 audit, $25K single workflow, $150K full network) signal a partner who has shipped enough to know what each shape of engagement actually costs.
3. Ownership transfer. Ask: "When we're done, what do we own?" The right answer is: code, prompts, architecture, documentation. The wrong answer is: access to a SaaS platform we operate. If the partner builds you a system that only runs while you're paying them, you've been locked in.
4. Stack opinions. Ask: "What stack would you use for X workflow and why?" A real partner has opinions formed by shipping the pattern multiple times. A pretend partner answers vaguely or says "whatever you want."
5. Honest scope. Ask: "What's NOT in scope for this engagement?" The honest partner names specific things they won't do (e.g., "we don't do change management, we don't do enterprise SOC 2 compliance buildout, we don't do full enterprise transformations"). The pretender claims to do everything.
The five red flags
1. Decks-as-portfolio. If their case studies are PowerPoint summaries instead of links to working systems, they're selling consulting hours, not implementation.
2. "We'll create a custom proposal." Translation: we don't know what this costs because we haven't done it before. Productized offers are a sign of expertise; blank-check pricing is a sign the opposite.
3. AI partnership badges as proof. "We're an OpenAI / Anthropic / Microsoft AI Partner" means almost nothing — partnership badges are sales tools, not expertise certifications. Every major AI consulting firm has all the badges. Doesn't tell you they've shipped anything.
4. Long sales cycles. If they require multiple discovery calls + a 4-week scoping process before quoting, the engagement will run the same way — slow + over-budget + scope creep.
5. They want to host / operate the system. "We'll run it for you" sounds friendly. It also means you don't own it and you'll pay forever. Real partners build, hand off, then retainer (optionally) for ongoing maintenance — but you own the system.
Questions to ask in the first call
Use these to filter signal from noise quickly:
On capability:
- "Show me 3 AI systems you've shipped that are running in production right now."
- "Walk me through the architecture of the most complex one. Why those choices?"
- "What's your default LLM provider, and why?"
On scope + pricing:
- "What's your fixed-price offer for a single workflow build?"
- "What's a typical engagement budget for a multi-workflow buildout?"
- "If I just want an audit, what does that cost and what do I get?"
On ownership:
- "When the engagement ends, what do I own?"
- "If you disappear in 6 months, can I still operate the system?"
- "Do you require an ongoing retainer for the system to keep running?"
On honest scope:
- "What's NOT in scope for your typical engagement?"
- "What kinds of clients are you a bad fit for?"
- "Tell me about a project you turned down. Why?"
A good partner answers these directly + specifically. A pretender answers them vaguely + with deflection.
How to structure the engagement
Once you've picked a partner, the right engagement structure for most SMB and mid-market buyers:
Phase 1: Audit (1-2 weeks, $4-10K). Strategic roadmap. Identifies which workflows to automate, in what order, on what stack. Worth doing even if you don't hire the same partner for the build.
Phase 2: First build (3-8 weeks, $15-50K). Single workflow end-to-end. Treats the first project as both productivity AND learning — you're also learning how to work with this partner + the methodology.
Phase 3: Expansion (ongoing). Additional workflows shipped one at a time. Usually a mix of build + retainer for maintenance + occasional new audit.
What to avoid: signing a "full transformation" engagement with no prior working relationship. The audit-first pattern lets both sides validate fit before committing real money.
Apply this to your business
Start with the $4,500 audit.
Pay, fill an intake, get a Claude Opus-written strategic roadmap inside a week. Async, no sales call.
Start the audit →Questions
How is Solidus different from "AI automation agencies"?+
Three things: (1) productized scope + pricing — published $4,500 / $25K / $75-150K offers, not custom proposals; (2) shipped production work — Morthn Intel, Teardown, the Moves feed are public, linkable, working systems; (3) ownership transfer — you get the code, the prompts, the architecture. We're built around the buyer experience most AI consultancies refuse to offer.
Should we hire a generalist agency or an AI-specialist team?+
Specialist. Generalist agencies that "added an AI practice" in 2024 are almost universally not deep enough to ship serious AI systems. Specialist teams that have been building AI products since GPT-4 launched are the only ones with real reps.
What about hiring an AI engineer in-house instead?+
For most SMB and mid-market: not yet. A single AI engineer costs $200-350K loaded and takes 6+ months to ramp on your business. A specialist team ships the first 2-3 workflows in that same window, at lower total cost, with shipped patterns from prior clients to draw from. Add the in-house engineer when you have 3+ workflows in production needing ongoing tuning.
Other guides
What Is AI Automation? A Practical 2026 Guide for Operators
9 min
The Real Cost of AI Automation (2026 Pricing Guide)
7 min
Build vs. Buy AI Automation: A Decision Framework
6 min
Which Workflows Should You Automate First? (Prioritization Framework)
6 min
The ROI of AI Automation: Honest Math (with Examples)
6 min