Insights top picks from 2025

Dear Readers, 

Thank you for your curiosity and engagement. The Insights editorial team extends gratitude to all Thoughtworkers, alumni, guests and clients who made Insights a hub for rigorous, practitioner-focused thought leadership.

As 2025 draws to a close, we're reflecting on a year where AI moved decisively from experimentation to execution.

The themes that resonated most with you revealed this shift: responsible tech frameworks became strategic imperatives, not aspirations. AI for software delivery evolved from automated coding to autonomous systems requiring greater engineering discipline. Data platforms transformed into federated ecosystems. And organizations learned that operationalizing AI means embedding it into decision systems while maintaining human oversight and transparency.

You engaged deeply with our technical explorations — context engineering, model context protocols, AI supply-chain transparency, and the infrastructure realities of managing AI workloads. The conversations around antipatterns like AI-accelerated shadow IT showed a maturing understanding: speed without discipline creates instability.

Operationalizing AI became the new frontier. Moving from innovation labs to production environments meant embedding AI into decision systems, workflows, and customer experiences — while maintaining transparency and human oversight. Your engagement with content on agentic workflows, multi-agent systems and AI-enhanced decision-making showed that organizations are realizing AI's value isn't just in automation, but in augmenting human judgment.

The five dimensions of AI adoption: May Xu believes organizations must move beyond a tools-only view for AI success. Her framework outlines five key dimensions for sustainable adoption: AI literacy, strong engineering practices, internal champions, sound AI governance (using labs and radars) and focused, use-case driven experimentation defined in an AI playbook.

How AI and a test-first approach can tackle modernization trade-offs: Irene Sandler and Shodhan Seth explore how the 'test-first modernization' approach breaks the traditional trade-off in mainframe migration. It uses AI to generate modern, idiomatic code against behavior captured from the running system, making complex rewrites safer and faster.

The 2025 DORA report: The 2025 DORA Report highlights that embedding AI in software engineering creates a 'productivity paradox,' boosting speed and quality but also increasing delivery instability. Thoughtworks argues that sustainable success requires a strong engineering foundation beyond simple tool adoption. Organizations must embrace systems thinking, leverage platform engineering to build in guardrails and prioritize developer experience. This shift means the engineer’s value moves toward prompt engineering and validating AI outputs, ensuring lasting, measurable value.

Humanizing AI strategy: Written by Tiankai Feng, author of Humanizing Data Strategy, this book shows you how to embrace adaptive AI governance, empower cross-functional teams to think through the ethical implications of AI, and, ultimately, create a culture of ethical innovation.

AI for software delivery evolved from automation to autonomy. The shift from "vibe coding" to context engineering marked a maturing understanding: AI doesn't replace developers — it redefines what engineering means. You explored goal-based development environments, neural delivery loops and the paradox that AI requires developers to think more strategically, not less. The enthusiasm around tools that reverse-engineer legacy systems and accelerate modernization reflected the urgent need to break free from technical debt.

Claude code saved us 97% of the work on the first try. then it failed utterly.: Alessio Ferri writes about an experiment using the Claude Code agent to accelerate language support in the CodeConcise tool yielded inconsistent results. For Python, it provided 97% of the necessary code in minutes; however, attempts to add JavaScript support failed completely. The results stress that while AI agents show huge promise, their output quality depends heavily on code quality, available libraries, and robust feedback loops

How much faster can coding assistants really make software delivery?: In this blog post, Sichu Zhang dismisses the claim that coding assistants increase delivery speed by 50%. Thoughtworks' testing, including a case study with GitHub Copilot, indicates that realistic cycle time improvements are typically 5% to 15%. This smaller, cost-effective gain is most realized when using AI for repetitive tasks and test generation.

Data platforms transformed into federated ecosystems. Centralized data lakes gave way to product-centric architectures where domains own and govern their data. The rise of data mesh adoption, synthetic data ecosystems and edge processing demonstrated that competitive advantage now hinges on how quickly and safely organizations can turn data into intelligence at scale.

Drive data and AI success with team topologies and data mesh: The complementary frameworks of Team Topologies (organizational structure) and Data Mesh (decentralized data management) are essential for modern data and AI success. Danilo Sato says by combining them, organizations can empower cross-functional data product teams, reduce bottlenecks, and accelerate innovation.

Data Quality ROI: This new book from Thoughtworker Gaurav Patole urges the world to approach data differently. Drawing on his experience helping organizations unlock long-term value from data, it combines practical wisdom with storytelling and bold conceptual thinking to provide readers with strategic clarity.

Responsible tech emerged as a non-negotiable foundation. With regulations like the EU AI Act entering force and global standards converging, organizations realized that governance isn't a constraint — it's an accelerant. Your interest in computational governance, policy-as-code and assurance frameworks showed that winning in the AI era means embedding responsibility into the fabric of how technology is designed and delivered.

Full throttle on data: Sebastian Werner says the upcoming EU Data Act should be viewed as an opportunity to unlock innovation, not a compliance burden. By mandating easier, fairer data access and sharing across devices, it facilitates new data-driven services and increased market competition, potentially boosting EU GDP by €270 billion. 

Besides these themes, software architecture and engineering practices continued to hold sway. For instance, this book on Facilitating sofware architecture by Andrew Harmel-Law was one of our popular reads. The book demonstrates that bridging the gap between architect and developer is possible. Laying down the conceptual underpinnings for a better approach and outlining practical steps that teams can take today, it opens up a new route to overcoming a key problem facing technology teams in every industry: architecture is just too demanding to be practiced by a single job role.

Technology Podcasts:

If you haven't heard the Thoughtworks Technology Podcast, you're missing out on getting a keen and deep understanding of trends that are shaping the software industry. Our technology podcast plunges deep into the latest tech topics that have captured our imagination. Join our panel of senior technologists to explore the most important trends in tech today. Get frontline insights into our work developing cutting-edge technology and hear more about how today’s tech megatrends will impact you. Some of our most popular episodes this year were:

Why the tech industry needs "expert generalists": In this episode of the Technology Podcast, Martin and Unmesh join hosts Prem Chandrasekaran and Lilly Ryan to discuss how they came to identify the importance of expert generalists and why it was important to not just talk about the issue, but to explicitly name it. They also explore how they believe the industry can cultivate and encourage expert generalists, despite an entrenched tendency to overlook their value.

We still need to talk about vibe coding: Vibe coding was, remarkably, named word of the year by the Collins English Dictionary at the start of November 2025 — pretty good going for a term that was only coined in February. We first discussed it on the Technology Podcast back in April, and, given its prominence in the collective lexicon this year, thought we should revisit and reflect on the topic as 2025 draws to a close.  To talk about it all and reflect on the implications, Thoughtworkers and regular podcast hosts Prem Chandrasekaran, Lilly Ryan and Neal Ford reconvened for a follow up to our April conversation. Taking in everything from the term's semantic slipperiness, its security risks and the challenges of maintenance, this is a discussion that, despite going deep into vibe coding, also touches on a huge range of issues in the technology industry today.

Context engineering: Tackling legacy systems with generative AI: On this episode of the Technology Podcast, Thoughtworks' lead for AI-enabled software engineering, Birgitta Böckeler, and tech principal Chandirasekar Thiagarajan join hosts Ken Mugrage and Neal Ford to discuss how it works. They explain the process, the tools and what the work is teaching them about both generative AI and legacy modernization.