How AI is Transforming Software Development — and What’s Holding It Back

I’ve been thinking a lot about how software development is evolving.

At some point in the future, thanks to AI, software development will be commoditized.

AI changes how humans add value. Tools like GitHub Copilot and Cursor or Windsurf are helping engineers write and debug code faster than ever. CodeRabbit is automating code review.

On the horizon, agentic systems like Devin are taking the lead—creating full outputs that engineers simply guide and refine.

Having used all of these tools, and having built out my own, I am noticing a roadblock that will prevent AI from taking the lead: context management.

AI tools can write code, generate plans, and review work, but they don’t understand your strategy, goals, or codebase history. This missing context is the gap we need to bridge.

---

The Evolution of Software Engineering

Using a Wardley map (shoutout to the Lethain for a recent post on this), I’ve drafted a visualization of how I see software engineering evolving from highly custom, manual work to commoditized, AI-led workflows.

1. Human-Led Engineering (The Past)

Historically, engineers wrote code line-by-line, crafted technical plans from scratch, and manually reviewed outputs.

• Example: Building an entire application by hand with minimal external tooling.
• Challenges: Slow, error-prone, and resource-intensive. Each piece of software was bespoke, with scaling limited by the size of engineering teams.

2. Human + AI Assistance Engineering (The Present)

Today, with significant and proper prompting, AI tools are now supercharging productivity by automating repetitive tasks and guiding decision-making.

• Examples:
  • GitHub Copilot: Suggests code snippets to accelerate development.
  • Cursor and Windsurf: Work alongside engineers in active editors.
  • Devin: Generates code independently and submits PRs for review.
• Impact: Engineers lead, but workflows are faster and more efficient. However, AI lacks the ability to adapt to specific company needs without significant prompting.

3. AI-Led Engineering (The Near Future)

Soon, agentic systems will take the lead in software engineering, generating outputs with minimal human input. Engineers will shift from executing tasks to overseeing and validating AI-generated work.

This is where the context problem becomes more critical. Without structured context, engineers will remain in a double-checking role, slowing adoption of AI-led engineering.

• Examples:
  • Define objectives, constraints, and priorities for AI.
  • Ensure outputs meet quality, technical, and business standards.
• Impact: Some teams are already using tools like Devin to create software solutions with minimal human input. Engineers review and refine the outputs, enabling faster iteration and delivery.

4. Engineering as a Service (The big shift)

Just as cloud computing shifted infrastructure from bespoke solutions to a utility, ‘Engineering as a Service’ will transform development into something similar.

Imagine a world where companies subscribe to platforms that generate entire applications tailored to their needs—just as easily as they use cloud infrastructure today. Engineers will no longer build systems from scratch; they’ll customize and extend pre-built solutions.

• Engineering as a Service: Companies might subscribe to platforms that generate end-to-end solutions with minimal customization.
• Impact: Engineers will focus on innovation and strategy, while routine development becomes commoditized.

---

Poor Context Management Blocks “Engineering as a Service”

AI today is like a Principal Engineer on their first day at a new company: they might be brilliant, but they know nothing about your strategy, processes, or systems.

To bridge this gap, AI needs access to structured and actionable context. Without it, AI-led workflows will remain out of reach.

---

A Translation Layer Converts Company Sources of Knowledge Into AI Accessible Context

The translation layer is the bridge that connects the potential of AI with the realities of your business, paving the way for commoditized software engineering.

1. Structuring Context: Package information like architecture diagrams, code documentation, and business priorities into actionable formats.
2. Dynamic Context Delivery: Provide AI with just-in-time access to relevant data, such as feeding architecture constraints during design or relevant source code when writing.
3. Bridging Human + AI Understanding: Train AI on company-specific norms and workflows, while curating outputs to align with team goals.

Today, some organizations are starting to experiment with custom solutions for context management, but most are still in the early stages. Examples include:

• Custom Vector Databases (e.g., Pinecone, Weaviate): Storing and querying embeddings to make unstructured data AI-accessible.
• Knowledge Graphs: Building structured relationships between company-specific entities (e.g., Neo4j or GraphQL-based solutions).
• Fine-Tuning LLMs: Customizing models on internal documentation and business data, though this is often resource-intensive.
• Knowledge Oriented Frameworks (e.g., RAGFlow, LlamaIndex): Creating services that dynamically provide relevant context to LLMs during tasks.

These approaches are primarily custom and require significant effort to integrate effectively. There are also emerging paid products attempting to tackle this problem, but none have fully addressed the complexities of managing business-specific context at scale. This represents a significant opportunity.

---

Context Unlocks AI-led engineering

By solving the context problem, we evolve towards:

1. From Assistance to Leadership:
   AI can independently generate codebases, plans, and tests with human oversight.
2. Standardized, High-Quality Outputs:
   Context-aware AI produces aligned results without constant re-prompting.
3. Commoditized Engineering:
   Routine tasks like code reviews and bug fixes become utilities, freeing engineers to innovate.

Strategically, solving the context problem represents a massive opportunity:

• Internal Acceleration: Organizations that build robust context solutions can leap ahead of competitors by unlocking faster, higher-quality AI-led workflows.
• Market Potential: As the demand for AI-led engineering grows, a universal, productized solution for context management could dominate a new market segment.

Currently, most context solutions are custom-built, with limited scalability or generalizability. This gap is ripe for innovation, and whoever gets it right, whether as an internal advantage or an external product, will define the next era of AI-driven development.

---

What This Means for Engineers

Early adopters of AI tools and context systems will lead the way in this transformation.

Software Engineers of tomorrow will look different than the Software Engineers of today:

• Engineers will shift from task executors to AI supervisors and innovators.
• Skills like architecture design, strategic thinking, and systematic problem-solving will become critical.

My advice for technology organizations thinking about how AI applies to them:

• Invest in AI Tools: Tools like Devin, Cursor, and Copilot are early steps toward AI-led workflows.
• Solve for Context: Build the systems that make business-specific context accessible to AI.
• Upskill Teams: Prepare engineers to collaborate with AI systems, not compete with them.

---

Closing Thoughts

Software engineering is on the brink of a revolution. Those who understand and act on the context problem today will shape how we all build tomorrow. The question isn’t whether AI will lead—but how ready we are to let it.

The companies that solve the context problem won’t just lead—they’ll define the next era of software engineering.