TL;DR: Technology strategy in 2026 requires disciplined portfolio thinking. Not everything new deserves investment, and not everything old deserves retirement. This post presents our opinionated tech radar – what to adopt with confidence, trial with intent, assess with curiosity, and hold with caution.
Gartner Projects $6.15 Trillion in Global IT Spending This Year – Where Should Your Budget Go?
Worldwide IT spending will reach $6.15 trillion in 2026, up 10.8% from 2025, with AI infrastructure alone accounting for $401 billion of new investment. In a landscape of relentless innovation and finite budgets, the CTO's hardest job is not finding interesting technology – it is deciding where to place disciplined bets and, equally important, where to stop investing.
Thoughtworks popularised the technology radar format in 2010 as a way to capture collective technology advisory opinion. The format endures because it solves a real problem: the four rings – Adopt, Trial, Assess, and Hold – provide a shared vocabulary for technology decisions that connects engineering reality to business strategy.
Here is ours for 2026, informed by client engagements, industry research, and hands-on delivery experience.
🟢 Adopt – Use With Confidence
These technologies and practices have proven themselves in production across multiple industries. If you are not using them, you are likely paying an opportunity cost.
Platform Engineering
Gartner forecasts that 80% of large software engineering organisations will have established platform teams by the end of 2026, up from approximately 55% in 2025 and 45% in 2022. Platform engineering has moved decisively from buzzword to business imperative.
The core idea: instead of every product team independently solving infrastructure, security, CI/CD, and observability, a platform team provides these as self-service capabilities through an Internal Developer Platform. Mature platform teams report 40–50% reductions in cognitive load for their developers and measurable improvements across all four DORA metrics.
Key tools: Backstage, Port, Humanitec, Kratix, Crossplane.
Our take: If you have more than five product teams and they are each solving the same infrastructure problems differently, platform engineering is not optional – it is overdue. See our detailed platform engineering analysis for implementation guidance.
OpenTofu
OpenTofu, the open-source fork of Terraform created after HashiCorp's BSL licence change in 2023, has reached production maturity. With strong backing from the Linux Foundation and contributions from Gruntwork, Spacelift, and env0, it is a credible – and increasingly preferred – alternative for organisations that value licence certainty.
Why adopt: MPL 2.0 licence certainty, feature parity with Terraform 1.6+, growing ecosystem support, and state file compatibility for straightforward migration. For most organisations, switching from Terraform to OpenTofu is a configuration change, not a rewrite.
Risk of inaction: HashiCorp's acquisition by IBM in 2024 has introduced additional uncertainty about Terraform's roadmap and pricing trajectory. OpenTofu provides a hedge at minimal migration cost.
Agentic AI
Agentic AI – systems that autonomously plan, reason, and execute multi-step tasks using tools – has moved from research to production in 2026. Gartner projects worldwide AI spending at $2.5 trillion this year. Frameworks like LangGraph, CrewAI, and AutoGen provide the scaffolding; foundation models from Anthropic, OpenAI, and increasingly capable open-source alternatives provide the reasoning capability.
Enterprise applications in production today:
- Automated code review and security remediation
- Customer service escalation handling with human-in-the-loop
- Document processing pipelines for legal and compliance
- Security incident triage and initial response
- Infrastructure troubleshooting and remediation
Critical caveat: Agentic AI requires robust governance. Autonomous systems need guardrails, audit trails, human oversight mechanisms, and defined blast radii. Adopt the pattern, but invest equally in the controls. Organisations deploying agents without governance frameworks are accumulating risk faster than they are generating value.
GitOps
GitOps – using Git as the single source of truth for declarative infrastructure and application configuration – is the default deployment model for Kubernetes-based workloads in 2026. Argo CD dominates, with Flux CD and increasingly native GitOps capabilities in managed Kubernetes platforms as alternatives.
Why it endures: Auditability through commit history, instant rollback via revert, drift detection and reconciliation, and developer familiarity with Git workflows. Organisations practising GitOps report significantly lower change failure rates and faster time to restore service.
FinOps as a Discipline
FinOps belongs firmly in the Adopt ring. With public cloud spend projected to exceed $1 trillion in 2026, FinOps adoption grew 46% in 2025, and 70% of large enterprises now maintain dedicated FinOps teams. The combination of the FOCUS standardisation, mature tooling – Kubecost, Infracost, Vantage, OpenCost – and organisational recognition that cloud cost management is a discipline rather than a project makes FinOps table stakes for any serious cloud operation.
Organisations using FinOps frameworks are 2.5× more likely to meet or exceed cloud ROI expectations. See our comprehensive FinOps guide for implementation detail.
🟡 Trial – Invest With Intent
These are ready for production use in specific contexts. Run deliberate experiments, measure outcomes, and build organisational experience before scaling.
WebAssembly Beyond the Browser
WebAssembly's server-side story has matured considerably. The WASI standard provides a portable, sandboxed execution environment that is faster to start than containers and more secure by default.
Use cases gaining traction:
- Edge computing via Cloudflare Workers, Fastly Compute, and Fermyon
- Plugin systems where users supply Wasm modules for extensibility
- Lightweight serverless functions with sub-millisecond cold starts
- Embedded policy evaluation – OPA already compiles to Wasm
Why Trial, not Adopt: The ecosystem is still developing. Debugging tools are limited compared to containers, library support varies by language, and the developer experience needs polish. But the fundamentals – near-native performance, sub-millisecond cold starts, and strong sandboxing – are compelling enough to warrant deliberate experimentation.
DORA Metrics at Scale
The four key metrics from the DORA programme – deployment frequency, lead time for changes, change failure rate, and time to restore service – are well-established as indicators of software delivery performance. Elite performers using DORA metrics are twice as likely to meet organisational goals, deliver faster customer value, and maintain higher developer satisfaction.
Why Trial now: The tooling to measure DORA metrics automatically has matured. Platforms like Sleuth, LinearB, Jellyfish, and native capabilities in GitHub and GitLab make measurement feasible without manual data collection. DX, Atlassian Compass, and Cortex also provide integrated DORA tracking.
Caution: Goodhart's Law applies ruthlessly. When metrics become targets, teams optimise for the metric rather than the outcome. Use DORA metrics for team-level reflection and improvement, not cross-team league tables or performance management. The moment you rank teams by deployment frequency, you have undermined the entire purpose.
Internal Developer Portals
Developer portals – centralised catalogues of services, APIs, documentation, and self-service capabilities – are the user interface of platform engineering.
Backstage dominates as a CNCF Incubating project, but alternatives offer different trade-offs:
| Portal | Strengths | Trade-offs |
|---|---|---|
| Backstage | Open-source, vast plugin ecosystem, CNCF-backed | Significant setup and maintenance effort |
| Port | Low-code configuration, fast time-to-value | Commercial, less customisable |
| Cortex | Strong scorecards and standards enforcement | Commercial, opinionated workflows |
| OpsLevel | Service ownership and maturity tracking | Commercial, narrower scope |
| Roadie | Managed Backstage – reduced operational burden | Vendor dependency on Backstage |
Why Trial: The value is clear for organisations with 50+ microservices. But implementation effort is significant, adoption requires cultural change, and the portal is only as good as the data and integrations behind it. Start with a focused use case – service catalogue, tech docs – and expand.
FOCUS Specification for Multi-Cloud Cost Normalisation
The FinOps Open Cost and Usage Specification standardises cloud billing data across providers. AWS, Azure, GCP, and Oracle have committed to FOCUS-compatible exports. For multi-cloud enterprises – 67% of enterprises now operate across two or more providers – FOCUS dramatically reduces normalisation effort.
Why Trial rather than Adopt: The specification reached 1.0 in 2024 and tooling is still catching up. Single-cloud organisations benefit less. But for multi-cloud environments, early adoption pays dividends as the ecosystem matures.
🔵 Assess – Watch With Curiosity
These are emerging and interesting but not yet ready for broad enterprise adoption. Track them, run proof-of-concepts if relevant, but do not bet strategy on them.
Quantum-Safe Cryptography
NIST finalised its first post-quantum cryptographic standards in 2024 – ML-KEM, ML-DSA, and SLH-DSA. HQC was selected as a backup standard in March 2025, with a draft standard expected in 2026. Cloudflare reports that although we will see the first post-quantum certificates in 2026, broad browser trust is unlikely before 2027.
The “harvest now, decrypt later” threat is real. Nation-state actors may be collecting encrypted data today with the intention of decrypting it when quantum capabilities arrive. NIST's IR 8547 draft lays out expected migration timelines, with a recommendation to deprecate vulnerable algorithms by 2030.
Our recommendation: Conduct a cryptographic inventory. Identify systems handling data with long-term sensitivity – healthcare records, financial data, government communications. Begin planning migration for those systems. Do not panic, but do not ignore it. The migration will take years; starting the assessment now is prudent.
Edge AI
Running inference at the edge – on devices, in factories, at cell towers – rather than in centralised cloud data centres. The hardware is ready: NVIDIA Jetson, Intel Movidius, Apple Neural Engine, and Qualcomm's AI Engine. The use cases are compelling: real-time quality inspection, autonomous vehicles, local language processing for privacy-sensitive applications.
Why Assess: Edge AI introduces operational complexity that most organisations underestimate. Model deployment and updates at scale, monitoring distributed inference, handling intermittent connectivity, and managing hardware lifecycle are all unsolved problems for most enterprises. The tooling for edge MLOps is immature compared to cloud MLOps.
AI Governance Tooling
Purpose-built platforms for AI governance – model registries with risk classification, automated bias detection, compliance reporting, and audit trails. The EU AI Act entered into force in August 2024, with full enforcement beginning in 2026. This creates regulatory urgency.
Vendors to watch: Credo AI, Holistic AI, IBM OpenPages, and governance extensions within MLflow and Weights & Biases.
Why Assess, not Trial: The market is fragmented and fast-moving. Many organisations are building governance capabilities on top of existing MLOps platforms rather than adopting standalone governance tools. Wait for consolidation, but track it closely given the EU AI Act enforcement timeline.
Rust for Infrastructure Tooling
Rust is increasingly appearing in infrastructure tooling: Cloudflare's Pingora HTTP proxy, the coreutils rewrite, and various CLI tools. Its memory safety guarantees and performance characteristics make it attractive for systems software.
Why Assess: For most enterprise teams, the hiring and training investment is significant. The ecosystem is maturing but remains smaller than Go's for infrastructure use cases. Worth assessing for performance-critical components – proxies, data-plane services, security-sensitive tooling – but not yet justified for general-purpose infrastructure work.
🔴 Hold – Proceed With Caution
These are not necessarily bad technologies. They are approaches where the industry has found better alternatives, and continued investment carries increasing opportunity cost.
Lift-and-Shift Cloud Migrations
Moving applications unchanged from on-premises to cloud VMs was an acceptable strategy in 2016 when the priority was data centre exit. In 2026, it is a recipe for high cloud bills, poor performance, and missed cloud-native opportunities.
The alternative: The 6 Rs framework – Rehost, Replatform, Refactor, Repurchase, Retain, Retire – applied deliberately to each workload. A mid-size enterprise that moved from blanket lift-and-shift to assessed migration typically saves 30–40% on cloud spend and gains access to managed services that reduce operational burden.
Monolithic CI/CD Pipelines
Single, centralised CI/CD systems – often Jenkins – managing builds for the entire organisation. They become bottlenecks, single points of failure, and configuration nightmares at scale.
The alternative: Decentralised CI/CD where each team owns their pipeline definition using GitHub Actions, GitLab CI, or Dagger, with platform teams providing shared components, templates, and golden paths. Consistency through convention, not centralisation. Jenkins served the industry well for fifteen years; in 2026, it is technical debt for most organisations.
Self-Managed Kubernetes Control Planes
Running your own Kubernetes control plane via kubeadm or kops when managed alternatives exist. The operational overhead of etcd management, API server availability, and control plane upgrades is rarely justified given the maturity of EKS, AKS, and GKE.
Exception: Air-gapped environments, specific compliance requirements, sovereign cloud mandates, or edge deployments where managed services are not available. Gartner projects sovereign cloud IaaS spending to reach $80 billion in 2026 – a 35.6% increase – so managed sovereign options are expanding rapidly.
Long-Lived Feature Branches
Feature branches that live for weeks or months before merging cause merge conflicts, delay integration testing, and undermine continuous integration.
The alternative: Trunk-based development with short-lived branches lasting hours to days, feature flags for decoupling deployment from release, and continuous integration that means what it says – integrating continuously. The data from DORA research is unambiguous: elite performers use trunk-based development and deploy multiple times per day.
Building Your Own Radar
A tech radar is a conversation starter, not a prescription. Every organisation's context is different. We recommend:
- Review quarterly – technology maturity changes faster than annual planning cycles can accommodate.
- Make it collaborative – the CTO should not author the radar alone. Involve senior engineers, architects, and delivery leads. Thoughtworks assembles a Technology Advisory Board; you should have your equivalent.
- Be honest about Hold – the hardest decisions are about what to stop, not what to start. Sunk cost bias is the enemy.
- Connect to investment – each Adopt and Trial item should have a sponsor, a team, and a budget. A radar without resource allocation is an aspiration document.
- Publish internally – transparency about technology direction reduces shadow IT and aligns teams. Make it accessible, not a slide deck that lives in a senior leadership folder.
Radar Maturity by Organisation Size
| Organisation Size | Recommended Approach | Review Cadence |
|---|---|---|
| 5–20 engineers | Lightweight – CTO maintains a shared document | Every 6 months |
| 20–100 engineers | Collaborative – architecture guild contributes | Quarterly |
| 100+ engineers | Formal – dedicated Technology Advisory Board | Quarterly with monthly updates |
What This Means for Your Organisation
Technology strategy in 2026 rewards disciplined portfolio thinking. The organisations that thrive are not those that adopt every new technology – they are those that make deliberate, well-informed choices about where to invest, experiment, watch, and divest.
Immediate Actions
- If you lack a platform engineering capability and have more than five product teams, begin planning now. The 80% adoption figure is not a trend to watch – it is the new baseline.
- If you are still on Terraform with concerns about the BSL licence, evaluate an OpenTofu migration. The switching cost is minimal; the risk reduction is significant.
- If you are deploying AI agents without governance frameworks, pause and invest in controls before scaling.
Medium-Term Investments
- Build DORA measurement into your delivery platform and use it for team reflection, not management reporting.
- Evaluate WebAssembly for edge and plugin use cases where container overhead is a constraint.
- Conduct a cryptographic inventory in preparation for post-quantum migration.
Strategic Positioning
- Watch AI governance tooling closely as EU AI Act enforcement begins.
- Resist lift-and-shift migration patterns for remaining on-premises workloads.
- Consolidate on GitOps and decentralised CI/CD if you have not already.
If you are building a technology strategy and want an outside perspective grounded in delivery experience – we are here to help.
