Make, Buy, or Automate

How The Intelligent Firm Wins In The AI Era

Hi there,

The rise of AI is creating firms that see their boundaries as a portfolio to optimise, not as walls to defend.

Firms used to exist to minimise transaction costs, but the disruption from APIs and AI drives transaction costs down. This new environment alters how firms structure and operate.

For example, Anthropic uses the agentic coding tool Claude Code to develop Claude Code and other tools, giving staff immense leverage.

If your developers aren’t using a tool like this, or if you aren’t enabling them to, you risk being outpaced by peers adapting to this change. For smaller tasks, with appropriate guardrails, an agent could manage bug fixes and minor features, while a senior engineer oversees code reviews as usual.

The New Cost Equation

The “theory of the firm” focuses on transaction costs. AI significantly lowers search, negotiation, and monitoring costs, reducing the challenge of optimising task-level decisions to make, buy, or automate.

Deploying AI incurs costs, with marginal costs measured in dollars per million tokens. A model's input tokens are typically cheaper than its output tokens. Google’s Gemini Pro 2.5 model costs $1.25 per million input tokens and $10 per million output tokens.

Additional costs of deploying AI include data storage and movement. Model training and maintenance can be costly, especially for fine-tuning specific needs.

This results in a new cost blend: transaction cost + computation cost. Intelligent firms leverage lower transaction costs but must avoid excessive computation costs to gain net benefits. They reduce salaries, vendor payments, and consultant costs, but now face material API token bills every month.

As transaction costs approach zero, computation costs become predominant. This redefines the “build vs. buy” logic and creates new opportunities for organisational design and firm structuring. The key question shifts to, “What tasks can I assign to a machine for an advantage?”.

The Triple Choice Framework

As AI models evolve, a new option beyond “make or buy” emerges: automation with AI agents. You can internally develop capabilities with AI tools, access them via APIs, or delegate problem-solving to AI agents.

The economics of making improve with lower marginal costs. If you have proprietary data, you can generate value. Additional benefits include control and customisation, which are vital for regulated industries.

In this AI landscape, some firms prefer building capabilities internally rather than buying from third parties, disrupting traditional business models. Consider how outsourcing firms, professional services, or legacy software vendors will compete.

Buying or outsourcing economics improve with an API-everything model, featuring low fixed costs and usage-based pricing. This allows flexible adjustments: use an API briefly, then stop.

However, reliance on API performance, quality, privacy, and security risks exists. According to several tech publications, Windsurf, recently acquired by OpenAI, lost its Anthropic API access with just 5 days’ warning. How can you mitigate that risk?

Automation with AI agents offers minimal human overhead, 24/7 operations, infinite scaling potential, and autonomous goal maximisation.

Yet, it demands skill and governance upgrades. Setting up workflows isn't enough; substantial planning, configuration, testing, and monitoring are critical.

This new environment means that decision-making itself becomes fluid. Optimal choices will shift daily based on task complexity, competition, and other factors, pushing firms to move from fixed to flexible operating models that adapt to demand.

Some tasks will require custom AI builds for migration, while others might temporarily use an API. As firms become more dynamic, monitoring or governing them will become complex due to shifting boundaries, rendering simplistic frameworks ineffective.

The intelligent firm transforms into a machine-centric entity, resembling a “dynamic firm” without meetings and emails; the logical endpoint is that it's all about API calls.

The Integration Paradox

This API-enabled modularity creates a strategic challenge: the integration paradox. Internal integration and unique data add value to non-commodity or secret information, potentially driving larger firms in specific sectors.

Internal data creates data synergy. These databases and records enhance each other: banks combine data, transactions, device data, and open-source intelligence to power algorithms that prevent fraud and reduce risk.

Competitive advantages arise from unique data and capabilities. Choosing between integration and microservices depends on the required capabilities. Do you think real-time processing is necessary? Should reinforcement learning be applied? Is it so unique that it has become the industry's “default choice?”

The push-everything-to-API approach works for well-defined tasks based on standards, where a “slightly better” version holds little value. Firms must consider how they deliver value to stakeholders.

Firms need to identify sustainable advantages in regulated industries with audit requirements. This shift will likely lead to more vertically integrated firms. Consider the contracts, service-level agreements, API specs, project documents, and communications from “@vendor.com” over the past decade. By analysing internal platforms, cloud services, APIs, and outsourced processes, they will find they have more data to create workflows than previously thought.

Becoming An Intelligent Firm

Turnaround times must be reduced to days. Being too slow can render your business obsolete.

Core competencies for an intelligent firm include boundary scanning, rapid reconfiguration, and portfolio management.

Boundary scanning means monitoring operations, revenues, and costs for near-real-time opportunity identification. Use AI to maintain a real-time intelligence system that tracks your industry and competitors, responding to threats within minutes or hours.

Rapid reconfiguration entails swiftly changing business practices by optimising make, buy, or delegate decisions, and leveraging low transaction costs for constant experimentation. Transition from A/B testing to continuous digital simulations, enabling extreme personalisation, so each target customer views a unique ad version.

Portfolio management focuses on creating frameworks for intelligent operations, directing AI agents toward optimal outcomes while eliminating blockers. This real-time risk management converts executive spaces into network-monitoring hubs.

Success is gauged by market share, speed of boundary changes, reconfiguration costs, error rates, retention from past experiments, and sustainable unit economic changes.

Choose one team and grant them unrestricted AI access; promote experimentation. Automate processes, utilise vibe coding, and empower top talent to innovate. Disrupt your organisation internally; otherwise, your best people will quit to start their AI-first startup.

The Competitive Reality & Your Response

AI-first startups achieve great results and rapidly attain high annual recurring revenue. AI wrapper firms struggle as frontier labs enhance functionality. Traditional firms lag due to slow progress. They’re not allowed to use AI coding tools.

Last year's B2B SaaS builders aim to apply AI-first thinking to legacy industries. "Boring" sectors hold immense economic potential. Professional services evolve, putting junior lawyers and entry-level roles at risk of obsolescence.

The opportunity cost is steep; soon, due diligence will question the necessity of specific roles. Firms that do not adapt may be seen as outdated. Under-investment in tech was acceptable in 2024 but will be critical by 2025 and beyond.

Boards must decide how quickly to shift to AI-first firms. Leaders progress, while followers strategise, and laggards see it as hype. Their failed cloud migrations make them distrust technology leaders, risking shareholder value destruction.

Some see this as mere hype, ignoring intense competitive pressures.

Companies often adopt specific tech platforms, such as IBM, Oracle, and Microsoft. This trend will extend to AI, with businesses favouring solutions like Anthropic, Gemini, or OpenAI to maximise efficiency. “Sign in with ChatGPT” will become standard corporate practice.

However, some industries will witness AI-first startups that automate and optimise processes from the start and require fewer resources than traditional tech teams.

The question is no longer whether this transformation is happening but whether you’ll embrace it or become a casualty.

Firms will soon be divided into two camps: the intelligent and the extinct.

What do you think?

Regards,

Brennan