The pitch is always the same: AI will do it faster. And honestly? For a lot of things, it's true. But the version that actually works in production consulting engagements looks nothing like the demos.
Here's what we've learned after integrating AI agents into our delivery workflow.
What Actually Changes
The biggest shift isn't code generation. It's the elimination of cognitively cheap work that used to eat senior engineers' time.
Boilerplate. Type definitions. Repetitive test cases. CRUD endpoints that follow obvious patterns. These weren't hard — they were just time-consuming. An experienced engineer could write a REST endpoint in 20 minutes, but those 20-minute tasks stack up fast across a sprint.
With agents handling pattern-recognition work, senior engineers spend their cycles on actual hard problems: distributed system design, security architecture, performance optimization, and cross-cutting concerns that require context no model has.
The Leverage Is in the Workflow, Not the Model
Every AI consulting pitch focuses on the model. GPT-4o. Claude Sonnet. Gemini. That's the wrong frame.
The real leverage is in how you integrate AI into your delivery loop. We've converged on a workflow that looks like this:
- Spec first. We write detailed specs before writing code. AI is good at generating code from specs; it's terrible at inferring intent from vague requirements.
- Agent-assisted drafting. For well-understood problems (an API endpoint, a data migration script, a UI component), we let agents draft and iterate quickly.
- Human review on all boundaries. Authentication flows. Database schema changes. External integrations. Security-critical paths. These get human eyes, always.
- Automated testing at the seams. We use agents to generate comprehensive test suites for the boundaries we care about most.
The result: we ship roughly 2–3x faster than a traditional team of the same size, without compromising on code quality where it matters.
What It Means for Clients
The most common client concern: "Are you just prompting an AI and billing us for it?"
No. Here's the honest answer: a junior developer following a spec can generate code quickly too. The value in a consulting engagement has never been raw code output — it's been the thinking behind the spec, the architecture decisions, the judgment calls under uncertainty, and the ability to debug what breaks in production.
AI makes the code generation layer faster. It doesn't replace the layer above it.
What it does change for clients is scope. Work that used to require a team of five can often be done with a team of two or three, operating at the same output level. That's a real cost reduction, and we pass it through.
The Skills That Still Matter
If AI handles so much of the implementation, what's left?
- Architecture design. Choosing the right storage technology, service boundaries, and deployment model requires experience and contextual judgment that current models don't have.
- Requirements engineering. Getting a clear, unambiguous spec out of a stakeholder is still a deeply human problem.
- Debugging complex failures. When something breaks in a way that crosses multiple system boundaries, models are helpful but not sufficient.
- Technical communication. Translating complex tradeoffs into language executives can act on remains a human skill.
These are the skills we've doubled down on as AI agents have absorbed more implementation work. It's not that software consulting is getting easier — the ceiling is just higher now.
Closing Thought
The consulting firms that will lose ground over the next five years aren't the ones that resist AI. They're the ones that use AI to produce more of the same thing: more junior developers, more hours billed, more marginal output.
The opportunity is to use AI to change the ratio — more senior thinking per dollar, less time-to-delivery, better outcomes. That's the bet we're making.
If you're evaluating a consulting partner, ask them how AI affects their team structure and pricing. The answer will tell you a lot about where they're headed.