Engineering the future
How AI is Transforming Engineering
by Jack Wang, August 2025
The Great Acceleration: AI's New Reality
AI has fundamentally accelerated both creation and scaling, compressing engineering velocity from months to hours. This new competitive landscape demands speed, as AI-native competitors are rapidly shipping solutions while others remain anchored to pre-AI processes. The cost of moving slowly now greatly exceeds the risk of imperfection.
Four Transformative Shifts Reshaping Engineering
Collaborative Renaissance
AI-facilitated collaboration collapses months-long development cycles into single sprints, turning strategic conversations into deployable solutions within hours.
Context Engineering
Engineers evolve from code writers to intent architects, mastering abstract patterns to create reusable mental models that scale across enterprise complexity.
Enterprise Intelligence
Your data flywheels, operational patterns, and institutional knowledge become an unbreachable competitive moat, compounding daily into living knowledge systems.
Brownfield Transformation
Legacy enterprises must transform rapidly. Culture shifting for speed is survival, leveraging decades of customer data and business logic before the window closes.
Part 1: The Dawn of Spec-Driven Engineering
Imagine a crisis, and within hours, AI transforms it into a production-ready solution. This is spec-driven engineering, where intent becomes reality at unprecedented speed.
1
7:49 AM: Crisis Strikes
A network outage causes call volume to spike by 500%, overwhelming support channels.
2
8:00 AM: Initial PRD Drafted
An AI assistant drafts a Product Requirements Document for an agent to guide customers through a 4G workaround.
3
9:00 AM: PRD Refinement
The PRD is updated with outage diagnostics, IVR flows, customer notes, and Openreach fixes, prepped for MVP deployment.
4
9:30 AM: AI Execution
AI agents execute the PRD, building prototype IVR flows, integrating escalations, and running tests.
5
12:00 PM: Pilot Beta Live
A pilot beta routes calls directly to the new IVR. Feedback begins pouring in by 3 PM.
6
3:30 PM: Frictionless Iteration
AI triages feedback, updates the PRD with urgent fixes, backlogs ideas, and sets up experiments for dynamic IVR solutions.
Vision → Velocity → Value. In a day.
EXECUTING THE PRD: FROM ENGINEERING CODE TO ENGINEERING INTENT
The Lossy Nature of Code
Traditional coding often leads to a disconnect between technical implementation and business value, losing critical intent in translation. Code-first approaches create misalignment and overlook deeper problem-solving.
Embracing Intent-Driven Engineering
With AI automating code generation, engineers must pivot to defining intent: specifying functional goals, constraints, and evaluation criteria. Machine-readable specs become the single source of truth.
Crafting Solutions with Mental Models
The rapid pace of AI demands deeper understanding, not just tool consumption. Engineers need adaptable mental models and blueprints to navigate complexity, ensuring robust, vendor-agnostic systems.
Heading Towards “Softwareless”: Feedback as the New Moat
With AI automating code creation, the true moat isn't proprietary software, but proprietary feedback. Every interaction – from a Netflix pause to a Tesla mile – generates invaluable intelligence.
While teams can now prototype systems in hours, the sustainable competitive advantage no longer lies in what you build, it lies in what you learn. AI democratises code creation, and it is the data supremacy creates unbreachable moats.
Future platforms will thrive by building "data flywheels," transforming every action into wisdom, securing an unbreachable competitive edge. This is defensive programming for the AI age.
Part 2: Welcome to Rome
Part 3: Standing on the Brownfield
Marshall Goldsmith's axiom haunts every leader: “What got you here won't get you there”.
For those standing atop enterprise brownfields, vast expanses of legacy systems, accumulated technical debt, and institutional inertia – this truth cuts deeper than any technological disruption before it.
You're now facing a new generation of developers who 'vibe code' applications into existence over weekends, whilst your organisation spends months in requirements gathering. You're watching startups with zero technical debt dance circles around your enterprise, whilst you're still filling out change request forms.
It’s no longer a matter of if transformation will happen, the moment for that has already passed. What remains is for you to determine how to take an active part in shaping the outcome, or risk being carried along by events outside your control.
The comfortable management playbook of waiting one human to pass on a message to another is becoming as obsolete as the COBOL systems still running in your data centres. What follows are uncomfortable truths that demand leader’s attention.
People: The Human Operating System Upgrade
Embrace the Inevitable Deskilling
Much of what you consider 'engineering expertise' is becoming commodity. But this isn't a tragedy, it's liberation. When AI handles the syntax and boilerplate, engineers can finally focus on what machines can't do well: understanding context, making judgement calls, and connecting technology to human needs.
Tap Into the Hive Mind
GitHub has become humanity's collective programming consciousness. Every problem you're facing, someone has already solved, or at least attempted. The best engineers aren't those who know the most, but those who can navigate this sea of knowledge most effectively.
Force the Culture Shift
Culture change in enterprises typically moves at geological pace. You don't have geological time. You have maybe 18 months before the AI-native competitors start eating your lunch, if it is not already happening.
Process: From Waterfall To Whitewater
Controlled Chaos Over Controlled Process
Your traditional frameworks were built for stability. In the AI era, constant change is the new normal, demanding adaptability.
Question Everything, Test Everything
Like engineers, critically evaluate every process. A task requiring 47 human approvals might be automated by AI in seconds, but only if you dare to test.
Bureaucracy Is Your Biggest Enemy
Governance designed for slow, six-month software releases will stifle AI innovation. These rigid processes, once necessary, now prevent rapid adaptation and growth.
Portfolio Thinking Over Project Perfection
Embrace a venture capitalist mindset: most AI experiments will fail, but the successful ones will deliver exponential returns. Speed and iteration outweigh striving for unattainable perfection.
Technology: Building For Constant Transition
Vendor-Neutral AI Platforms
Adopt an "AI Independence Architecture" using abstraction layers to treat hyperscalers as interchangeable utilities. Your competitive edge lies in the platform you own, prioritizing speed and flexibility.
Align Vision with AI Agents
Craft an AI vision where agents orchestrate seamless customer experiences, portable across providers. The competitive moat is your unique business context and logic, not the underlying foundation models.
Data Flywheels & Intelligence Sovereignty
Build self-reinforcing data flywheels, transforming interactions into intelligence. This proprietary cycle of learning from customer behavior is an unduplicatable asset, securing true AI sovereignty.
Transition as the New Steady State
Embrace constant change. Legacy systems are anchors; focus on vendor-neutral abstraction layers allowing every component to be replaceable within 90 days. Fast iteration is the new normal.
Build Fast, Pivot Fast
The cost of being wrong has plummeted; the cost of being slow has skyrocketed. Start new AI solutions in isolation, prove their value, and then manage scaled transitions. Avoid endless migration planning.
Learn Fast, Sustain Intent
Maintain crystal-clear business intent, even as implementations change daily. Version your specifications like code; they are your institutional knowledge and portable intellectual property, impossible to reverse-engineer.
SOMEWHAT RADICAL IDEAS
Cultivate the Generation-Verification Loop
Stop protecting obsolete skills. Empower your teams to rapidly audit and guide AI output as part of context engineering. This ability to refine AI is the new engineering superpower.
Mandate English as the New Programming Language
Every team member, technical or not, needs an active GitHub presence. The new programming language isn't Python, it's clear, precise English, enabling effective AI interaction and specification.
Create 'AI Friday' Policies
Suspend traditional processes and require teams to accomplish goals using AI tools. Make the discomfort productive. Those who adapt become evangelists; others reveal themselves as transformation blockers.
Establish 'Innovation DMZs'
Create protected spaces where teams can experiment without triggering every compliance alarm. Set clear boundaries: customer data stays protected, but everything else is fair game for rapid prototyping.
Build AI Communities of Practice
These aren't committees; they're working groups solving real problems with AI tools. When marketing uses AI for campaign variants and finance automates reconciliation, culture shifts organically.
Institute Process Archaeology
Map every approval step, document requirement, and sign-off process. Ask: "Could an AI agent do this in seconds?" If yes, eliminate the human bottleneck. Reward process elimination.
Made with