Should security teams build or buy their tools?
Security teams should buy front-line controls such as EDR, firewalls and email security, and build the data and automation layer above them on a governed platform with SSO, approvals and audit. Censuswide research for Dimension by ThreatAware found 77% of UK security teams have built internal tools with AI coding assistants, but 69% have had tools delayed or scrapped over governance concerns.
Rob Jardine8 min read
Buy the tools that act on threats, and consider building the ones that join your data together. Front-line controls such as endpoint detection and response (EDR), firewalls and email security are usually best bought, because building them takes specialist engineering. The layer above them, which turns data from those tools into visibility, reporting, automation and control, is often worth building, because its value comes from fitting your own systems and processes, and AI coding assistants have made it far quicker to build.
The catch is production. An application built in an afternoon still needs hosting, sign-in, permissions, approvals and an audit trail before it can be trusted with live systems. That’s why many teams now take a third route: build the application themselves, and run it on a platform that already provides those foundations.
In research commissioned by Dimension by ThreatAware and conducted by Censuswide (opens in a new tab), 77% of 300 UK IT security decision makers, at businesses whose security teams build their own tools, said their teams had used AI coding assistants to build internal security tools in the past 12 months. But 69% said tools had been delayed or scrapped because of security, governance or compliance concerns.
Why has the build vs buy question changed for security teams?
The build vs buy question has changed because AI coding assistants let security teams build in hours what once took months of development. Until recently, building your own security tooling meant a development project: engineers, a backlog and months before anything was usable. So most teams bought, even when a product only half fit.
A pattern ThreatAware, the company behind Dimension, sees often is a team that sits through a few vendor demos, then builds its own version with an AI coding assistant instead.
Josh Thomson, ThreatAware’s CTO, describes the practical limit of what a team can build as a frontier that is “shifting outward, at pace”. His example: two years ago, few would have attempted to build an MFA management tool even for a small startup. Now a large enterprise can tackle MFA with an application built with AI.
Which security tools should you buy, and which can you build?
Security teams should usually buy the front-line tools that act on threats, and consider building the layer above them that connects those tools together. This split is adapted from Josh Thomson’s article Introducing Dimension (opens in a new tab).
| Front-line tools | The layer above them | |
|---|---|---|
| What they do | Act directly on threats: detect, block, filter and scan | Join data from other tools to give visibility, intelligence, automation and control |
| Examples | EDR (such as CrowdStrike Falcon or Microsoft Defender), firewalls, email security, vulnerability scanners | Access reviews, coverage reports, compliance trackers, exposure dashboards, custom automations |
| Buy or build? | Usually bought. Building them takes specialist engineering, from firewall hardware to email authentication standards such as SPF and DKIM. | Often buildable. Each team decides case by case what to buy and what to build. |
Josh’s “complexity-squared” test is a useful guide:
- A simple product in a complex environment can be built.
- A complex product in a simple environment can be built.
- A complex product in a complex environment is usually better bought.
When is it worth building your own security tools?
Building your own security tool is worth it in three situations: to join systems together, to replace a tool you keep for a few features, or to get a tool you can’t buy.
- To join systems together. Combining data from identity, device, security and HR tools gives context no single tool has, and lets you act on it.
- To replace a tool you keep for a few features. Many security products are only partly used. Josh’s rule of thumb, based on the Pareto principle: “80% of customers use just 20% of the capability, despite paying for 100% of it.”
- To get a tool you need but can’t buy. Some tools don’t exist off the shelf, and others are out of reach on budget or headcount.
What does it take to run a homegrown security tool in production?
A homegrown security tool needs hosting, sign-in, permissions, approvals, an audit trail and an owner before it can be trusted with live systems and other people’s data. Building the application is now the easy part; running it safely is not. In full, it needs:
- Hosting and a database, with backups, patching and monitoring.
- Sign-in through your single sign-on (SSO) and user provisioning.
- Roles and permissions, so each person sees only the data they should.
- Approval before any change to a live system, such as disabling an account or changing a configuration.
- An audit trail your assessors can query.
- Encrypted credentials for every system it connects to.
- Version control, so changes can be reviewed and rolled back.
- Network isolation and compliance with your own policies.
- An owner who can maintain it when the person who built it moves on.
Many homegrown applications stall at this stage, or ship without these controls. In the same research, 45% said self-built tools don’t go through the same review and approval process as other software, and 41% had no formal policy for AI-built internal security tools, or didn’t enforce one.
Two problems follow. The first is ownership: a tool works well until the one person who understands it is unavailable, and nobody else can safely change it. The second is sprawl: one application becomes many, each built differently and none of them audited.
How do buying, building and building on a governed platform compare?
Buying gives you a finished tool, building gives you full control, and building on a governed platform gives you that control with the production foundations already in place. Here is how the three compare, as of November 2026.
| Buy off the shelf | Build it yourself with AI coding assistants | Build on a governed platform (such as Dimension) | |
|---|---|---|---|
| Data | The vendor’s integrations and data model | You write and maintain every connector, including API limits and changes | Connect each system’s API once into a shared, read-only data layer |
| Governance | SSO, roles and audit usually included | You build SSO, roles, approvals and audit, or ship without them | SSO, roles, approvals and a tamper-evident audit log included |
| Building with AI | Limited to the vendor’s configuration options | Fast to prototype with tools such as Claude Code, OpenAI Codex or GitHub Copilot | Build in plain language, or from your own AI coding agent over MCP |
| Deployment | Ready on day one | You provide hosting, a database, patching and monitoring | Hosted for you, with isolated tenancy, backups and version control |
| Changes | Wait for the vendor’s roadmap | Change anything, if the builder is still around | Change in plain language, in shared workspaces others can maintain |
| Pricing | License fee, often for more than you use | Low to start; ongoing engineering time, hosting and AI usage | Platform subscription (Dimension from $49 per month (opens in a new tab)) plus time to build |
| Suits | Front-line controls, and teams with no capacity to build | Personal tools, or teams with engineers to own security and upkeep | Team tools on your own data that need to reach production |
When should you buy, build, or build on a governed platform?
Buy front-line controls, build personal tools yourself, and build on a governed platform when a tool joins data from several systems and a team will use it.
Buy an off-the-shelf tool when…
- It’s a front-line control, such as EDR, a firewall or email security.
- A product already fits most of what you need, and you’ll use most of it.
- You need it working on day one and have no capacity to build or maintain anything.
Build it yourself when…
- It’s a personal or one-off tool that doesn’t touch live systems or other people’s data.
- You have engineers with time to own hosting, security and maintenance for the life of the tool.
- Policy requires it to run entirely inside your own infrastructure.
Build on a governed platform when…
- The tool joins data from several systems you already own.
- More than one person or team will use it, with different levels of access.
- It needs to change live systems, such as opening tickets or disabling accounts.
- It has to pass a security review or an audit.
What should you ask before building a security tool?
Before building a security tool, ask what it does, who will use it, what it changes, who will maintain it and whether it would pass a security review.
- Is this a front-line control, or does it join data from tools you already have?
- Will anyone other than the builder use it?
- Will it change anything in a live system?
- Who will maintain it in a year, if the builder has moved on?
- Could it pass your own security review today, with SSO, permissions, audit and approved changes?
If it joins data, other people will use it and it changes live systems, the build is worth doing, but it needs a production foundation from day one rather than as a hardening project at the end.
Where does Dimension fit in build vs buy?
Dimension fits between buying and building from scratch: it’s ThreatAware’s governed workspace for security applications, where teams build applications on their own data with the production foundation already in place. That foundation includes SSO and role-based access on every plan, a tamper-evident audit log, a person’s approval before any action changes a live system, and isolated tenancy on AWS.
- Data (opens in a new tab): connect your systems through their APIs, such as your identity provider, device management, endpoint protection, HR and ticketing tools, into one live, read-only data layer.
- Studio: describe the application you need and AI Composer builds it on your live data.
- Actions: let applications open tickets, send messages or change configurations, with a person approving each action.
Teams that already build with AI coding agents can keep doing so. Dimension connects to tools such as Claude Code, Claude Desktop, VS Code and OpenAI Codex over MCP, so an application built there runs in production behind SSO, with approvals and audit in place.
Teams that would rather not start from scratch can begin with a ready-made application from the App Library (opens in a new tab), such as a user access review (opens in a new tab), an MFA coverage report (opens in a new tab) or a Cyber Essentials readiness check (opens in a new tab).
Dimension doesn’t replace front-line security tools. It connects to them through their APIs, like any other system.
Related
Frequently asked questions
Whether building or buying a security tool is cheaper depends on what you count. Building with AI coding assistants is cheap to start, but the ongoing costs are engineering time, hosting, AI usage and the security work needed to run it safely. Buying has a predictable license cost, but you often pay for features you don’t use, so compare the cost over the tool’s whole life, not just the first version.
AI coding assistants can build a security application quickly, but they don’t provide the production foundation on their own: hosting, SSO, permissions, approvals, audit and someone to maintain it. That’s where many homegrown tools stall. In Dimension’s research with Censuswide, 69% of UK IT security decision makers said tools had been delayed or scrapped over security, governance or compliance concerns.
Most security teams shouldn’t stop people building tools with AI, because banning it tends to push building out of sight. A better approach is a clear policy plus a governed place to run what people build. In the same research, 41% had no formal policy for AI-built internal security tools, or didn’t enforce one.
When the person who built a homegrown tool leaves, the tool often stops being maintained, because nobody else understands how it works or can safely change it. Tools built in shared workspaces, with permissions and version control, can be picked up and changed by the rest of the team.
Whether the AI sees your data depends on the tool. An AI coding assistant running on your own machine can usually see the data and query results it works with. In Dimension, models are never trained on your data, and AI Composer works from the shape of the data rather than the values.
Dimension connects to your systems through their APIs, including identity providers such as Microsoft Entra ID or Okta, device management such as Microsoft Intune or Jamf, endpoint protection such as CrowdStrike Falcon or Microsoft Defender, HR systems such as Workday, and ticketing tools such as ServiceNow or Jira.
Dimension doesn’t replace your existing security tools: front-line tools such as EDR, firewalls and email security are usually best bought. Dimension connects to them and to your other systems, so you can build the applications that sit across them.
About the author
Rob Jardine is Chief Marketing Officer at ThreatAware (opens in a new tab), the cybersecurity software company that builds Dimension. ThreatAware was founded in 2018, is ISO/IEC 27001:2022 certified and provides security software to more than 100 enterprise organizations across Europe and North America.
Sources
| Statistic or fact | Source |
|---|---|
| 77% used AI coding assistants to build internal security tools; 69% had tools delayed or scrapped; 45% skip the usual review; 41% have no enforced policy | Research commissioned by Dimension by ThreatAware and conducted by Censuswide: 300 UK IT security decision makers at UK businesses with 1,000 to 10,000 employees and a security team that builds its own tools, excluding the public sector. Data collected 16–22 October 2026. Coverage in Cyber Protection Magazine (opens in a new tab) |
| The two-layer split, the complexity-squared test, the 80/20 rule of thumb and the “frontier” quote | Josh Thomson, Introducing Dimension (opens in a new tab), 23 October 2026 |