Based on Andrew Ng's "The Four Ps" in The Batch (Oct 23, 2026): https://www.deeplearning.ai/the-batch/the-four-ps

AI is changing how engineering teams build software, but giving developers access to coding assistants is not enough to transform an organization. The real opportunity comes from redesigning how people work, which projects receive attention, what infrastructure supports development, and how decisions move through the company.

For engineering leaders, this requires more than introducing another tool or encouraging employees to experiment. It means creating an environment where AI becomes part of the team's operating model, not simply an optional addition to traditional workflows.

The transition can be understood through four connected priorities: People, Projects, Platforms, and Processes. Each addresses a different obstacle, and neglecting one can limit the benefits of the others.

An AI-native organization starts with people who understand how to work effectively with AI.

Buying subscriptions to coding assistants does not automatically make developers proficient. Engineers need practical experience using AI to understand unfamiliar code, generate implementations, investigate bugs, write tests, and evaluate technical alternatives. More importantly, they must recognize when AI output is unreliable and know how to verify it.

This requires a learning culture where experimentation is encouraged rather than treated as a distraction from regular work. Teams need time, appropriate tools, and opportunities to explore how AI can improve their specific responsibilities.

Engineering leaders should participate in this process themselves. A manager who has never worked directly with coding agents may understand their advertised capabilities but still struggle to judge their limitations. Building even a small application, experimenting with an agent, or deploying an AI-powered feature provides firsthand experience that can improve technical and business decisions.

That experience also changes the questions leaders ask. Instead of simply asking whether AI can generate code faster, they can investigate where it saves meaningful time, what review is required, and which tasks still demand substantial human judgment.

For existing engineering organizations, the goal is not necessarily to replace the entire team with people who already specialize in AI. Developing current employees can create a more sustainable transition because they already understand the company's systems, customers, and technical history.

Tip: Give engineers dedicated time to experiment with AI on real tasks, then encourage them to share useful workflows, failed attempts, and verification techniques. Leaders should also build something themselves so their decisions are informed by practical experience rather than assumptions.

AI experimentation is useful, but experimentation alone does not guarantee business results.

Developers often discover promising applications by exploring problems from the ground up. Someone might automate a repetitive workflow, improve an internal tool, or find a faster way to investigate production issues. These discoveries can reveal opportunities that would never emerge from a top-down strategy meeting.

However, promising experiments need leadership support if they are going to become meaningful products. A prototype may demonstrate that something is technically possible, but turning it into a dependable solution requires resources, ownership, integration, and a clear understanding of its intended users.

This is where engineering leaders play an important role. Their technical knowledge can help the organization distinguish between an interesting demonstration and an opportunity worth pursuing.

A common mistake is treating AI primarily as a cost-reduction initiative. Automating repetitive work and improving engineering efficiency can deliver value, but the larger opportunity may be creating products and services that were previously too expensive, slow, or difficult to build.

For example, an organization might use AI to reduce the time employees spend searching internal documentation. That is useful, but the same underlying capabilities could support a customer-facing assistant, improve product discovery, or provide a new service that creates additional revenue.

The distinction matters. Efficiency improvements help an organization perform existing work better. Growth-oriented projects can expand what the organization is capable of offering.

A balanced AI strategy should therefore include both operational improvements and experiments that could create new customer value. Not every experiment will succeed, so projects should be evaluated through clear milestones rather than enthusiasm alone.

Tip: Maintain a pipeline of AI experiments and select a few with strong potential for broader impact. Evaluate them based on the problem they solve, user demand, feasibility, measurable outcomes, and the resources required to scale them.

Even capable AI agents cannot perform effectively when they lack access to the information and tools required to complete a task.

An agent working on a software feature may need more than access to a repository. It may also need product requirements, architecture decisions, technical documentation, previous discussions, support tickets, and relevant test results.

Without that context, an agent can generate technically plausible code that conflicts with existing design decisions or overlooks important business requirements.

This makes platform infrastructure a central part of AI adoption. Organizations need to prepare their systems and information so agents can retrieve relevant context and perform authorized actions safely.

Several areas deserve attention:

Security and governance must be built into this foundation. Giving an agent access to more information or the ability to take action should not mean giving it unrestricted permissions. Confidential data, production infrastructure, customer information, and critical systems require appropriate access controls and safeguards.

For some organizations, the first obstacle may be more basic: developers are still prohibited from using AI tools in their regular workflows. Establishing clear, secure policies that allow responsible AI usage can be an important first step.

The objective is to make the environment useful for agents without compromising the systems and information the business depends on.

Tip: Start by identifying the information and tools developers repeatedly need to complete common tasks. Make those resources accessible to approved agents through controlled permissions, clear documentation, and secure execution environments.

AI can reduce the time required to implement clearly defined software requirements. But faster implementation does not automatically make the entire development process faster.

When coding becomes less expensive, other activities can become the limiting factors. Teams may spend more time deciding what to build, clarifying requirements, reviewing generated changes, coordinating dependencies, or validating whether a solution actually addresses the problem.

This changes where engineering effort should go.

In traditional workflows, developers may receive a detailed specification and focus primarily on implementation. In an AI-native environment, engineers can contribute earlier by helping shape product requirements, evaluate possible solutions, and identify technical constraints before development begins.

That broader involvement becomes valuable because producing code is no longer the only major challenge. Understanding the right problem and making good decisions can have a greater influence on the final result.

Organizations should therefore create teams with enough context and authority to make meaningful decisions without waiting for unnecessary approval at every step.

High-context teams understand the product, users, technical systems, and business priorities. High-empowerment teams can act on that knowledge within clearly defined boundaries.

This does not mean removing all oversight. Significant risks such as confidential data exposure, major service outages, and permanent data loss still require safeguards. The goal is to apply controls where the consequences justify them rather than slowing every routine decision through excessive approvals.

Coordination also becomes more important as multiple teams adopt AI agents. If agents work independently without sharing relevant information, they may create conflicting changes, duplicate effort, or overlook dependencies between systems.

Shared documentation, clear ownership, common standards, and emerging ways for agents to exchange information can help keep development aligned.

Tip: Review the entire development process instead of focusing only on coding speed. Identify where work gets delayed, give teams the context to make decisions, and establish safeguards for high-impact actions without creating unnecessary approval steps.

People, Projects, Platforms, and Processes are not separate initiatives that can succeed independently.

Skilled engineers need promising projects where they can apply their abilities. Those projects need platforms that provide reliable information, tools, and secure execution environments. And even the best technology will struggle to deliver results if outdated processes prevent teams from making decisions or coordinating their work.

Consider a team that has adopted AI coding assistants but receives incomplete requirements. Developers may generate code faster, yet spend the saved time clarifying specifications and correcting misunderstandings. If the team also lacks access to relevant documentation, the agents may produce changes that do not fit the existing system.

The problem is not necessarily the AI tool. It is the operating environment around it.

An effective transformation addresses these connections together. It develops people, funds projects with real potential, prepares the technical foundation, and redesigns workflows around the capabilities and limitations of AI.

An AI-native engineering team is not simply a traditional team with coding assistants installed. It is a team that understands how to combine human judgment with machine capabilities across the entire development process.

Engineers still need technical foundations, product understanding, communication skills, and accountability. AI can help them move faster, but it cannot remove the responsibility to choose the right problem, verify the result, and protect the systems they build.

For engineering leaders, the next step is to look beyond tool adoption. Examine whether people are developing the necessary skills, whether projects are creating meaningful value, whether platforms give agents the right context, and whether processes allow teams to act effectively.

The organizations that make progress will not necessarily be those using the most AI tools. They will be the ones redesigning their engineering environment so people and AI can produce better outcomes together.

Hit reply and tell me: which of the four Ps is your team's weakest right now, and what's your first move? I read every answer.

Keep going, one small spark at a time.

Erwin