Tech Radar
What we would use on your project today.
Our working position on enterprise technology, updated quarterly from delivery experience. The Hold ring is the useful one.
Q3 2026
Four rings, one opinion each
Positions reflect what we have run in production for clients. Where we have not, we say so and place it in Assess.
Proven across multiple engagements. Our default choice.
OpenTelemetry
Vendor-neutral instrumentation is now mature enough to be the default on every new service.
Infrastructure as code with policy gates
Terraform or Bicep with OPA/Checkov in CI. Drift becomes a build failure rather than a discovery.
dbt for warehouse modelling
SQL-first transformation with tests and lineage that analysts can review without learning a new language.
Trunk-based development
Short-lived branches with preview environments per pull request. The single largest lead-time improvement available.
Passkeys for workforce authentication
Phishing-resistant, and adoption friction is now low enough for general rollout rather than pilot.
Working well in production, but not yet our unconditional default.
Retrieval-augmented generation on governed corpora
Effective where the source data is already curated. Value drops sharply when it is not.
Platform engineering with golden paths
Strong results at scale, but needs a real platform team — otherwise it becomes another abandoned internal tool.
eBPF-based runtime security
Excellent visibility with low overhead. Operational maturity varies considerably between vendors.
Temporal for long-running workflows
Removes a great deal of bespoke orchestration code. Worth the learning curve above a certain complexity.
Lakehouse as the single store
Increasingly viable as a replacement for separate lake and warehouse, with caveats around concurrency.
Promising. We are running controlled experiments, not client work.
Agentic workflows in production systems
Compelling demonstrations; evaluation and rollback design are still the unsolved parts for regulated use.
WebAssembly outside the browser
Genuine promise for plugin isolation and edge compute. Tooling is still early for enterprise delivery.
Confidential computing
Attractive for regulated multi-party data, though the operational overhead is still significant.
Post-quantum cryptography migration
Inventory your cryptographic estate now. Migration timelines are longer than most organisations assume.
We would need a specific reason to choose these on new work.
Microservices as a starting architecture
A modular monolith first, with services extracted where a boundary has proven itself, is cheaper and usually faster.
Heavily customised packaged suites
Customisation outside supported extension points converts every vendor upgrade into a project.
Manual evidence collection for audit
If control evidence is assembled by hand before an audit, the control is not actually continuous.
Long-lived feature branches
Merge pain scales super-linearly with branch age. Almost every symptom blamed on 'the codebase' starts here.
Reading the radar
What each ring commits us to
The rings describe what we would do on your engagement, not an abstract opinion about the technology.
| Ring | What it means | What we will do on a client engagement | Evidence required to move in |
|---|---|---|---|
| AdoptDefault choice | Proven across multiple engagements, we would default to it | Use without a conversation | Shipped and operated in production for over a year |
| Trial | Working well, but not yet across enough contexts | Propose it, with the trade-off stated | Shipped in at least one production engagement |
| Assess | Promising, actively investigated in the lab | Not on client work unless you ask and accept the risk | A prototype measured against a stated bar |
| Hold | We would argue against starting something new with it | Maintain existing use, do not extend | Moving out requires a documented reason |
Adopt
Default choice- What it means
- Proven across multiple engagements, we would default to it
- What we will do on a client engagement
- Use without a conversation
- Evidence required to move in
- Shipped and operated in production for over a year
Trial
- What it means
- Working well, but not yet across enough contexts
- What we will do on a client engagement
- Propose it, with the trade-off stated
- Evidence required to move in
- Shipped in at least one production engagement
Assess
- What it means
- Promising, actively investigated in the lab
- What we will do on a client engagement
- Not on client work unless you ask and accept the risk
- Evidence required to move in
- A prototype measured against a stated bar
Hold
- What it means
- We would argue against starting something new with it
- What we will do on a client engagement
- Maintain existing use, do not extend
- Evidence required to move in
- Moving out requires a documented reason
Questions
How to read this without over-reading it
Particularly the 'Hold' ring, which people reasonably misinterpret.
How the radar works
Is 'Hold' the same as 'bad'?
No. Hold means we would not start something new with it, which covers mature technology that is being superseded, tools with a shrinking hiring market, and things that are perfectly good but wrong for the kind of work we do. Plenty of our clients run Hold technology very successfully.
How often does it change?
Quarterly, and items move slowly by design. A radar that reshuffles every month is tracking fashion rather than experience. Moving something into Adopt requires it to have been in production for over a year, which is deliberately a high bar.
Will you use something that is not on the radar?
If you already run it, yes — the radar describes what we would choose, not what we can work with. If you are choosing fresh and want something in Assess or absent entirely, we will say what we do not know about it and let you make the call with that stated.
Disagree with a position?
We would genuinely like to hear it — several entries have moved because a client argued us out of them.