Build or buy: agentic platforms vs a delivery consultancy

By Nick Beaugeard · 7 minute read · ← All posts

Two years ago "AI coding tools" meant GitHub Copilot, full stop. Today the landscape looks different. Cursor, Devin, Factory, Lovable, Windsurf, Claude Code, Replit Agent, v0, Bolt — all pitch some flavour of "autonomous software engineering." On top of that, every consultancy — us included — now has a named agentic delivery engine.

Founders and mid-market CIOs are rightly asking a simple question: do I buy one of these platforms, or hire a consultancy with one of its own, or both? Here is how we'd think about it.

What the platforms are good at

Standalone agentic coding platforms are extraordinary at one specific thing: compressing the individual engineer's inner loop. A developer with Cursor writes more code, better, and faster than the same developer without Cursor. Lovable and v0 get a working web prototype in front of a stakeholder in an afternoon. Devin can take a well-scoped ticket and produce a pull request.

When the bottleneck is "we have capable engineers, they need to ship more", a platform is the right answer. It is cheap compared to headcount, it is measurable, and it ships improvement in weeks.

What the platforms are not good at

Platforms are weak at the work upstream of the engineer. They do not, on their own:

• Decide what to build.
• Weigh the architectural trade-offs that make or break a product.
• Negotiate scope with a stakeholder who is not sure what they want.
• Sit in a board conversation about risk.
• Take commercial accountability for the outcome.

That work requires a senior human, and the platform cannot substitute for it any more than a great IDE substitutes for a great architect.

A simple decision framework

Run your engagement through four questions:

1. Do you already have capable senior engineers? If yes, platform-led is probably the right call. Buy Cursor for the team, add a senior design system for prompts and reviews, and ship.
2. Do you know what you are trying to build? If yes, platform-led works for execution. If "what we are trying to build" is itself unresolved, platform-led will produce clean code for the wrong thing.
3. Is your target stack mainstream? Platforms do best with React, Next.js, Python and well-trodden stacks. If your target is .NET on Azure with specific enterprise integration patterns, a consultancy that knows the stack cold will outperform a generic platform.
4. Can you run the engineering review culture internally? Platforms produce plausible code. Plausible code is not production code. Without someone in-house who can review, refactor, and reject, platform output becomes tech debt at machine speed.

Four common shapes

Matching the answers produces four common engagement shapes we see in the market:

A. Internal senior team + platform

You have a CTO, a couple of seniors, and a clear product direction. You don't need a consultancy; you need Cursor for the team and a bigger Azure OpenAI quota. Buy, don't hire.

B. Consultancy + their delivery engine

You have a brief, a decision-maker, and no engineering team. You need both the senior thinking and the compressed delivery loop. Hire a consultancy with a credible agentic engine (ours is Symphony; there are a handful of real ones in the AU market). You are paying for judgement and output, not tools.

C. Consultancy + your platform

You have engineers already using Cursor or Claude Code, but a specific product needs senior architectural direction and pace. Bring in a consultancy to lead the work, use their delivery engine for the senior's output, and let your internal team operate within the consultancy's framing. This is the pattern we see most in mid-market modernisation.

D. Platform + staff augmentation

You have direction but not bench. You do not need a consultancy; you need more hands using the platform you've chosen. Staff aug plus platform is the pragmatic answer. We would not be the right firm for this shape.

Where Released fits in this framework

Released is specifically shaped for B and C — engagements where senior judgement plus a credible delivery engine is the value. Symphony is the engine; the principal is the judgement. If your engagement looks like A or D, we will tell you so in the first call, and point you at the tools or staffing you actually need.

Buying a platform makes your engineers faster. Hiring a senior makes your product better. Most real engagements need both.

The category is called "agentic AI development and ongoing services" for a reason. The "agentic" and the "services" are not substitutes. The interesting firms — platforms and consultancies alike — are the ones that are honest about which half of that they actually do.