What is the Bitter Lesson?
Your team is six months deep into building a custom AI solution. Next week, a new foundation model makes it obsolete with a simple prompt. Welcome to the Bitter Lesson.
The “Bitter Lesson” was articulated by AI researcher Rich Sutton in an influential 2019 essay reflecting on 70 years of AI work. Sutton, a pioneer in “reinforcement learning” or RL (the final stage of training for popular models like ChatGPT) observed a consistent pattern: complex approaches to AI consistently lose to simpler methods that become feasible with more compute power.
Foundation models have certainly proven this point. They are the very epitome of vast compute power distilled into a simple “transformer” neural network. Each wave seems to have taught itself new superpowers just by being larger than the one before, incrementally obsoleting generations of carefully crafted software investment and human expertise.
The lesson is “bitter” when one of these commoditized models obsoletes IT infrastructure that is still a work in progress, triggering write-offs and sunk costs where ROI and competitive differentiation were expected. In the tech community, this phenomenon is summarized as “AI is eating software,” meaning AI will continue to grow its role as a provider of the capabilities we have traditionally crafted in code.
The core insight is that as computing power grows exponentially (Moore’s Law), AI gets smarter. Systems that scale with that AI talent will keep getting better, while systems that put constraints or dependencies on current AI capabilities will hit a ceiling. Ultimate winners will harness the growing depth AI with an evergreen strategy for newer models, and guardrails for operating safety.
explaining why advances feel so rapid.
Why This Matters for Your AI Strategy
The Business Implications
Your carefully crafted AI solutions may become obsolete faster than you expect. What took your team many months to build might be baked into the next off-the-shelf AI model.
The accelerated rate of AI advancement creates major challenges for enterprises:
- Planning cycles can’t keep up: Traditional enterprise planning cycles (12-24 months) are now longer than AI capability cycles (3-6 months). By the time you finish building a comprehensive AI solution, the underlying assumptions may no longer be valid.
- Tech investments become obsolete: Heavy investments in proprietary AI infrastructure, custom models, or specialized frameworks carry increasing risk of obsolescence. The more bespoke your approach, the more vulnerable it becomes to being leapfrogged by general-purpose AI advances.
- Solutions are difficult to abandon: Teams can become emotionally and professionally invested in complex solutions they’ve built, making it psychologically difficult to abandon them for simpler approaches, even when the simpler approaches deliver better results.
Under the Bitter Lesson, your competitive advantage increasingly comes not from having the most robust AI architecture, but from being the fastest to adopt and deploy new AI capabilities as they emerge.
For example, think back to when companies used to spend years building complex chatbot frameworks with decision trees, intent recognition, and elaborate conversation flows. Now, a basic prompt to GPT-4 often delivers better customer service with zero custom development. The companies that pivoted quickly gained advantages over those still maintaining their custom systems.
4 Strategies to Minimize the Bitter Lesson’s Impact
Given that your next AI initiative could be leapfrogged by advances happening right now, how do you minimize the risk of building something that’s outdated before it launches? The answer isn’t to stop innovating. It’s to fundamentally change how you innovate.
The companies that will thrive aren’t those building the most complex AI architectures. They’re the ones that can swiftly capture value from each wave of AI advancement while staying flexible enough to pivot when better approaches emerge.
1. Prioritize Speed to Business Value
Don’t build the “perfect” AI architecture or chase the latest capabilities. Focus on solving real business problems that matter and deliver results quickly in short cycles. Prioritize ROI and measurable outcomes, not impressive technical solutions.
Example: Instead of building a comprehensive AI strategy platform, start by solving your high-pain operational challenges. Get measurable savings within 90 days, then tackle the next problem.
2. Design for Technical Agility
Build systems that can easily swap between different AI models and approaches. Avoid getting locked into specific platforms or APIs that might become obsolete.
Example: Create API abstractions that let you switch between local models, OpenAI, Anthropic or Google according to what your business problem needs at the time, without rewriting your entire application.
3. Test Against Your Own Reality, Not Generic Standards
Create custom evaluation criteria based on your specific business needs. What works for others might not work for your context.
Example: Rather than choosing AI models based on published benchmarks, test them against your actual customer inquiries and measure accuracy on your specific use cases.
4. Embrace Change as the Only Constant
Accept that your current AI approach will become obsolete. Plan for it. Be willing to abandon previous work when better approaches emerge.
Example: Budget for AI solution refreshes more frequently, rather than expecting your current implementation to last multiple years.
Turn the Bitter Lesson into Your Strategic Advantage
AI’s exponential growth has fundamentally shifted the competitive landscape, handing early movers a powerful advantage. The Bitter Lesson is unavoidable, but understanding it means you can develop solutions that capture value through each major AI wave.
While your competitors invest months in custom solutions destined for obsolescence, you’ll be leveraging each breakthrough as it emerges. While they debate long-term architectures, you’ll be generating measurable returns. While they’re caught off guard by the next model release, you’ll be ready to capitalize.
Your advantage starts today: Identify your most pressing operational challenge and solve it with available AI tools. While others get trapped by complexity, you’ll be setting the pace in your industry.
