Find the AI tools nobody approved
Skip to content
Blog

Not all security tools are created equal

Honey, I Shrunk the Stack!

Dan Horscroft6 min read

Hundreds of vendors, one headache

The security market keeps growing. New vendors appear every year, categories split into sub-categories, and in a world of acquisitions, tools often stretch well beyond their core focus. EDR platforms add exposure management, SIEMs add automation, identity tools add posture scoring. Two tools in your stack can claim the same job, and a third can do half of it. Nobody’s quite sure who signed off the fourth.

It’s easy to lose sight of what each tool is actually there to do. In this blog we want to look at them with a different mindset, because we think that underneath all the category confusion sits a much simpler picture.

Act or connect - A new lens

For years, the industry has organised security around categories. Analysts define them and vendors compete within them, so budgets tend to get built around them too. That works well for selling tools, but it isn’t always the best way to run a security programme.

We think there’s a better way to look at it. Forget what a tool is called and look at what it actually does, and you’ll find nearly every tool lands in one of two layers.

Layer 1 acts on your environment. Everything that happens in your stack starts here.

Layer 2 connects what Layer 1 knows. Everything you learn from it comes together here.

That might sound obvious, but it changes the conversation. Instead of asking “what kind of tool is this?” you start asking “what job is this doing for us, and who should be doing it?”

Some tools will straddle both. XDR acts and correlates, big platforms bundle a bit of everything, and an AI SOC agent might pull a Layer 1 lever on its own. That’s fine. The point isn’t to file every product in the right drawer, but to know which jobs you’re actually paying for.

This idea isn’t completely new. Back in 2021, Gartner’s Cybersecurity Mesh Architecture (CSMA) described something similar, with distributed controls working together through shared layers of analytics, identity, policy and dashboards. What’s different now is who can build those shared layers. When CSMA was introduced, building them yourself wasn’t realistic for most teams. AI has changed that, and it raises a fair question about who should own them.

If there’s one shift we’d encourage security leaders to make, it’s to stop managing a list of products and start managing two layers.

Layer 1 - Leave it to the specialists

The EDR, the firewall, the email filter. Identity providers, SASE, MDM. These are hard problems, solved by specialists.

A good EDR is years of threat research, kernel engineering and detection tuning, tested against attacks seen across thousands of environments. An MDM vendor has spent years keeping pace with every OS update, device model and enrolment quirk that Apple, Google and Microsoft throw at it. That expertise is really what you’re buying, and for now it’s well beyond what most internal teams could replicate.

Nobody builds their own EDR. At least, not yet. (If someone on your team has volunteered, keep an eye on them. They may be on to something.) For now, Layer 1 should probably stay bought, and for good reason.

Layer 2 - Making a phenomenally complicated world manageable

Every Layer 1 tool produces its own stream of alerts, logs, telemetry and findings, in its own format, with no idea how your organisation works. Layer 2 is where that changes. Alerts get triaged against what matters to you, findings turn into a prioritised plan, compliance gets evidenced, and you and the board get an answer you both understand.

It gives you visibility, intelligence, automation and control over everything Layer 1 does, through things like SOAR, the AI SOC, GRC, exposure management and reporting. It’s also where most of the market lives, and where it overlaps the most.

What makes Layer 2 different is that it’s mostly a data problem rather than a deep engineering one, and the tools to solve data problems are now in most security teams’ hands.

That doesn’t cover all of Layer 2. The SIEMs and security data lakes underneath it are serious engineering problems in their own right, and threat intelligence is only as good as the number of organisations it sees, which no single team can match. Most teams will keep buying those. Some large organisations with deep engineering benches will build even these, but for everyone else the bigger opportunity is on top, in the workflows, logic and context that make that data useful to you.

Why building is now on the table

Until recently, building your own Layer 2 tooling meant finding developers and waiting your turn in someone else’s backlog. Security teams had the domain knowledge but not the engineering capacity, so they bought.

That barrier has largely gone. AI coding assistants let an analyst describe what they need and get working code back. Every major Layer 1 vendor publishes an API, and the knowledge of how to use it is a prompt away. Vibe coding has turned “we’d need a developer for that” into an afternoon’s work.

The people who understand the problem can now build the solution. A SOC analyst can automate their own triage, or a security engineer can build the report the board actually asked for, without waiting on anyone.

Building still has a cost. The script that took an afternoon to write can take years to maintain, usually starting the week its author leaves. That trade-off is what most Layer 2 decisions now come down to.

Building is easy now. Production is not. Anything your organisation relies on needs access controls, an audit trail and someone signing off before it changes a live system. That’s the gap between a clever script and a proper tool.

We’ll come back to that later.

Thinking in layers

Once you see your stack this way, tooling decisions get a lot simpler.

Build or buy becomes a better conversation. It usually comes down to four questions:

  1. What is the tool for? Look past the category and the marketing. What does it actually do, and who relies on it day to day? The answer usually tells you which layer it belongs in.
  2. What problem does it solve? Get clear on the problem before you look at solutions. Is it something you’re dealing with today, and is anything already solving it?
  3. What is it worth? Look beyond the licence and think about the time it saves, the risk it reduces and the effort it takes to run. What would it cost to go without it?
  4. Do we need to buy it? Could you build it, extend something you already own, or live without it? This is the answer that’s changed most in the last few years.

Overlap becomes visible. Three Layer 2 tools reporting on the same Layer 1 data is a consolidation opportunity, not a coverage strategy. More often than not, it’s also three dashboards that disagree with each other.

Vendor value deserves a harder look. A Layer 1 vendor earns its place through expertise you can’t easily reproduce. For a Layer 2 vendor, it’s worth asking what you’re paying for that your team couldn’t build, whether that’s speed, maintained integrations, patented technology, scale or just time back. Sometimes that justifies the spend and sometimes it doesn’t.

So, build or buy?

The market will keep sprawling, but most tools will still either act or connect.

Our advice is to leave the acting to the specialists. Layer 2 is different, because it’s where your organisation’s context, priorities and processes live, and building it is now well within reach. So, before you sign off the next tool on your shortlist, ask which layer it sits in and what it would take to build it yourself.

If the answer is build, remember that production problem from earlier? This is something we’ve been working on with Dimension. It’s a place to build your own Layer 2 tools on top of the data you already have, with the entire production side taken care of. Right down to keeping things running when the author moves on.

Build or buy, we’d love to hear how your team draws the line.

Build it. Approve it. Run it.

Try Dimension free for 14 days on your own data. No credit card, nothing to install, and nothing changes in live systems without human approval.