I love the mindset of building. Whether it's the skill, the invariable grit needed or just the joy of creating something that fits exactly what you need.

We’re building at Vigilant AI.ai and it’s one of the key things that attracted me to the team.

So the question is not build or buy broadly speaking. It is…

when does “build your own” become a problem?

Or, the simpler question I’ve learned to ask myself in many situations over the years (some will disagree):

what’s the benefit?

In my experience, the risks sit in three buckets: speed, expertise, and cost vs. value. In AI, those risks are amplified because the pace of change is relentless. A large programme can be out of date before it goes live. There is real value in letting others fail, learn, and ship the foundations you can stand on.

1) Speed

Internal platforms take time. AI does not wait.

While committees debate architecture and control frameworks, competitors ship on managed rails. Models improve, prices shift, vendors release safer defaults. By the time a big internal build reaches daylight, the world has moved on and you are already on version minus one.

Watch for: long dependency lists, “phase two” safety work, a launch date that keeps slipping, and a pilot that looks good but never becomes the default path in the workflow.

What to do instead: pick one frontline moment, land a named teammate in that flow, and prove movement on a line metric within 90 days.

2) Expertise

Standing up a model is easy. Running a safe, reliable service is the job.

Real production means trusted data paths, retrieval from approved sources, policy-aware prompts, evaluation, red teaming, monitoring, decision logs and clear handoffs when humans must decide. That needs people with time to care for it. In AI, the bar rises every quarter as practices mature.

Watch for: plans that lean on heroics, thin coverage for risk and audit, and no dedicated time for evaluation or drift.

What to do instead: reuse proven bricks for safety and observability, then put your effort into the thin layer that solves the job to be done.

3) Cost versus value

The expensive part is not compute. It is people and time.

Every sprint spent wiring registries and monitors is a sprint not spent improving a customer outcome. If your build does not live inside everyday work with a clear measure and sensible guard rails, it will not move the number that matters.

Watch for: business cases that quote capacity, not outcomes; roadmaps that celebrate platform milestones instead of adoption and results.

What to do instead: measure one number the function already cares about, publish it weekly, and keep shipping tiny changes that move it.

What buying actually gives you

• Pace with a safety baseline. Mature products ship trusted retrieval, policy-aware UX, decision logs and live monitoring by default. You start from a safer place and get to outcomes faster.
• Leverage from others’ learning. Let vendors and the wider market eat the cost of failed ideas and edge cases. You benefit from their updates without paying to rediscover them.
• Focus. Your best people work on the parts customers feel, not on plumbing you can rent.

When you should build

Keep it simple. Build when:

• It gives you a clear edge others cannot buy.
• Law or policy means your data and controls must sit inside tight boundaries.
• You have a specific performance or integration need that a managed option cannot meet.

Even then, mix approaches. Buy proven building blocks. Build the thin layer that makes you different. That pattern lasts.

A little about Vigilant AI.ai

We deliver AI teammates for regulated businesses. We enable productivity, safely, with real-time guardrails.

We believe the future of work is AI teammates collaborating with humans to lift outcomes. Others share that belief. Where we differ is how it comes to life:

• Do it in the flow. Real effectiveness happens inside your existing systems, not in yet another app.
• Protect in real time. Protection should prevent issues as work happens, not sit in a log after the fact.
• Empower teams. Give people the tools to shape teammates that solve real problems in their context.

We encourage leaders to see AI differently. Stop treating it like software. Treat it like a teammate. Like any new hire, it needs onboarding and coaching, and people need time and evidence to trust it before it reaches peak productivity.

‍