From Geek to Star #46 - Emerging Engineering approaches in the AI-age

If you missed the previous episodes, you can access them online here.

🗓️ This Week – Episode 45: engineering practices and organisations are profoundly changing and we are starting to see the effects. 

I had the opportunity to share with you how Sau Sheong Chang, CTO of Singapore GovTech (the governmental agency spearheading the digital and AI transformation of Singapore government and public services) envisions the future of engineering in this age of AI. In newsletter #41 “The unwritten job description”, I shared one of his papers sharing his thoughts on the shift in how engineering organisations build and operate. In newsletter #42 “enabling AI in your organisation as a tech leader (⅔)”, I shared another of his papers on how “context engineering” and “harness engineering” are also fundamentally changing how to do software engineering. 

A few days ago, I held an inter-company session to hear from the founder and CEO of a digital native startup which has been around for 10 years. In January 2025, this founder, an engineer by background, decided to use himself AI coding assistants to develop some features he was frustrated were taking too much time. He was then so impressed that he decided to equip all his company with GenAI assistants and to transform how the company operated and the roles, skills and mindset. 

Reflecting on my session with the digital native founder and the insights from Sau Sheong Chan at GovTech, I see a clear comparison between a simple shift to a “digital startup” mindset and the more profound evolution into an “Agentic Engineering” model in complex organisations like big traditional companies. Both share the foundational philosophy of using AI as a massive productivity multiplier, but the Agentic framework in a complex organisation demands a much deeper structural and economic transformation.

Common emerging engineering practices

• The Inversion of Bottlenecks: Both models agree that coding speed is no longer the primary constraint. AI generates code instantly. This means the new bottlenecks are human decision speed, a lack of imagination, and legacy governance/approval processes designed for human-speed delivery.

• The Evolving Engineer: The human role is completely redefined. Whether you call them an outcome-based “Builder” or a “Supervisory Engineer,” the job is shifting from writing syntax to exercising strong judgment, writing precise specifications, and constantly evaluating/correcting AI outputs.This require though to have strong technical oversight capability, which raises the question of whether you can have this quality if you never had to struggle to build yourself complex systems. Something I am still pondering regarding how the young graduates can build up their value in this age of AI.

• Cultural Resistance & The Generational Divide: Both highlight that senior personnel often resist this transition because it challenges their professional identity as implementers. We can't just passively hand out tools. Both recommend structured interventions, like 1-on-1 coaching or protected experimentation time.

• The Absolute Need for Guardrails: Both a small organisation size to a big one warn against “vibe coding” without safety nets. We must understand that AI perfectly amplifies existing organizational dysfunction. Success requires aggressive automated testing and embedding compliance and security directly into the deployment pipeline.

Key Divergences: How complexity models the approach

While the commonalities show a shared starting line, it appears to me that the size of the organisation, or at least its complexity (due to legacy, history…) brings fundamental differences in approach:

• The “Builder” scope: in a traditional SDLC approach, different phases are managed by different persons (ideation / design / specifications / development / testing / production). What is costing time and money is the handover and loss of context from one phase to the other. Smaller digital native organisations are shifting to people able to handle different phases leveraging on AI, thus drastically reducing the entropy of the traditional way. In bigger traditional companies, this model is a very good inspiration and can probably be applied in some areas, while others may still require more specialised people (still augmented with AI) given the complexity and dependencies to apprehend. 

• The Value of Code (Ephemeral vs. Asset): in a smaller digital native setup, one may view code as the final product, just delivered X times faster by leaner teams. For those thinking Agentic Engineering, however, the argument is that because agents can regenerate code instantly from instructions, the source code becomes disposable (ephemeral). The durable artifacts an organization must truly preserve are system specifications, domain models, and decision histories. For larger and traditional organisations, it seems to me that even being into agentic engineering, this last approach is more achievable as many in a large company may not have the skills to interact with the code to understand what the systems do. One way or the other, AI is however there to be able to accelerate and make it much easier to keep the knowledge up to date if well designed.

• Technical Formalization (Testing vs. Context/Harness): Instead of just basic prompting and CI/CD, as systems become more complex and interdependent, teams must rely on two formal disciplines, as Sau Sheong Chan shared: Context Engineering (rigorously designing the system instructions and data) and Harness Engineering (the infrastructure that constrains and orchestrates the agent's behavior).

• Economic & Market Disruption: This model enables a “solo-to-scale” model where a single person can build a proof-of-concept without a team. For digital native companies which are used to have only internal staff, this is a great advantage. However, larger companies used to massively outsource their delivery and use of off the shelves solutions can not benefit of this approach. This actively threatens the business models of outsourced IT vendors and makes custom-built AI software a viable alternative to expensive SaaS subscriptions, which was not the case before.

• Infrastructure Strategy: for large companies with many systems, this mandates that organizations must build centralized, enterprise-grade platforms like shared tool,  registries, and agent deployment pipelines to safely scale from operating one agent to thousands.

The Takeaway:

Transitioning to an AI-driven organization is the first step, providing faster execution and leaner teams. Smaller digital native organisations have already been actively forging their way with impressive outcomes, something tech leaders in bigger corporations can take inspiration from.  However, in larger, more complex organizations, achieving scalable, safe, and disruptive AI transformation requires a paradigm shift driven by complexity.

This mandates: 

• Shifting the focus from valuing the code itself to preserving system specifications and domain models with a structured knowledge management approach for both humans and AI - leveraging on AI to do so; 

• Formally engineering the AI’s environment through Context and Harness Engineering; and building centralized, enterprise-grade platforms to safely orchestrate thousands of agents. 

• Rethinking fundamentally the approach on internal vs outsourcing, on make vs buy as custom-built solutions enabled by AI is a real alternative now.

🙏 I’d Love to Hear From You

Has your company started to move in that direction? What do you think helps to do the first next step in the right Direction?

Reply to this email, I read every note.

Follow me on LinkedIn for more reflections and “behind-the-scenes” thinking between newsletters. Don’t hesitate to comment or reshare,  it’s one of the best ways to grow your SHINE 🌟. If you want to know more about how I can support you or your teams to thrive in a tech career in this AI-age, have a look at my offerings here.  

P.S. Referral Pilot 🚀

Forward this email to one engineer or tech friend who could also benefit from this newsletter: sharing is caring - a little gesture can go a long way to strengthen bonds.

✨ May the SHINE be with you!

From Geek to Star by Khang | The Way Forward