𝗠𝗼𝘀𝘁 𝘁𝗲𝗮𝗺𝘀 𝗳𝗼𝗰𝘂𝘀 𝗼𝗻 “𝘂𝘀𝗶𝗻𝗴 𝗔𝗜”

However, my interest lies in whether 𝗔𝗜 𝗶𝘀 𝘁𝗿𝘂𝗹𝘆 𝗱𝗲𝗹𝗶𝘃𝗲𝗿𝗶𝗻𝗴 𝗯𝘂𝘀𝗶𝗻𝗲𝘀𝘀 𝗼𝘂𝘁𝗰𝗼𝗺𝗲𝘀.

Utilising AI tools is straightforward, but delivering real value with AI is where it becomes intriguing.

---

𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴 & 𝗣𝗿𝗮𝗰𝘁𝗶𝗰𝗲 #𝟭

When engineers adopt tools like GitHub Copilot or Cursor, they primarily change how they write code. Yet, when an organisation embeds AI into its core business architecture, a deeper transformation occurs:

👉 Traditional engineering practices begin to break down.

This is where an extended skillset becomes essential—regardless of the terminology we use.

---

𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴 & 𝗣𝗿𝗮𝗰𝘁𝗶𝗰𝗲 #𝟮

The real shift is from:
𝘋𝘦𝘵𝘦𝘳𝘮𝘪𝘯𝘪𝘴𝘵𝘪𝘤 𝘴𝘺𝘴𝘵𝘦𝘮𝘴 → 𝘪𝘧 𝘟, 𝘵𝘩𝘦𝘯 𝘠
to
𝘗𝘳𝘰𝘣𝘢𝘣𝘪𝘭𝘪𝘴𝘵𝘪𝘤 𝘴𝘺𝘴𝘵𝘦𝘮𝘴 → 𝘪𝘧 𝘟, 𝘵𝘩𝘦𝘯 𝘱𝘳𝘰𝘣𝘢𝘣𝘭𝘺 𝘠.

In discussions with senior tech leaders, one thing is evident:
Managing that uncertainty at scale is not merely a coding problem.

It evolves into:

• A systems design challenge
• A testing and validation challenge
• A cultural shift in our perception of quality.

---

𝗘𝘅𝗮𝗺𝗽𝗹𝗲: 𝗧𝗲𝘀𝘁𝗶𝗻𝗴 𝗣𝗿𝗼𝗯𝗮𝗯𝗶𝗹𝗶𝘀𝘁𝗶𝗰 𝗦𝘆𝘀𝘁𝗲𝗺𝘀

One area of exploration is how testing must evolve.
We transition from binary correctness to evaluation-based systems (“evals”):

• Optimising for statistically significant improvement over time
• Accepting variability while maintaining quality control.

In practice:

• Using LLM-as-a-Judge patterns to assess outputs (factuality, helpfulness, tone)
• Running multiple models in parallel to validate outputs
• Maintaining golden datasets and conducting regression-style evals on every change.

---

Where I currently stand

Across most areas of the AI SDLC, a hybrid model appears most effective:

• Deterministic where precision and control are vital
• Probabilistic where flexibility and adaptability add value.

The challenge lies not in choosing one approach but in designing systems that effectively balance both.
This remains a working hypothesis that I am actively refining. However, the direction is becoming clearer:

👉 For AI-native delivery (beyond just AI-assisted coding), we must rethink how we design systems, define quality, and evolve engineering practices.

---

Discussion

I’m curious about how others are navigating this:
𝘈𝘳𝘦 𝘺𝘰𝘶 𝘭𝘦𝘢𝘯𝘪𝘯𝘨 𝘵𝘰𝘸𝘢𝘳𝘥𝘴 𝘥𝘦𝘵𝘦𝘳𝘮𝘪𝘯𝘪𝘴𝘵𝘪𝘤, 𝘱𝘳𝘰𝘣𝘢𝘣𝘪𝘭𝘪𝘴𝘵𝘪𝘤, 𝘰𝘳 𝘥𝘦𝘴𝘪𝘨𝘯𝘪𝘯𝘨 𝘧𝘰𝘳 𝘢 𝘩𝘺𝘣𝘳𝘪𝘥 𝘮𝘰𝘥𝘦𝘭?
💬𝘋𝘔 𝘮𝘦—𝘐’𝘮 𝘦𝘢𝘨𝘦𝘳 𝘵𝘰 𝘤𝘰𝘮𝘱𝘢𝘳𝘦 𝘯𝘰𝘵𝘦𝘴 𝘸𝘪𝘵𝘩 𝘰𝘵𝘩𝘦𝘳𝘴 𝘵𝘢𝘤𝘬𝘭𝘪𝘯𝘨 𝘵𝘩𝘪𝘴 𝘴𝘩𝘪𝘧𝘵.