Amazon Warns Engineers: Skip 'Bleeding Edge' AI Tools

Key Takeaways

• Amazon prioritizes 'working, effective solutions over cheap ones' and will optimize compute costs later
• Engineers are told to use the best approach for each problem, whether that involves LLMs or not
• Teams must bring domain expertise to AI pilots rather than expecting AI teams to learn their area

Amazon has codified its internal approach to AI development with a set of six engineering principles that read more like a reality check than a hype document. The guidelines, obtained by Business Insider, tell engineers to avoid constantly chasing the latest AI advancements and to prioritize solutions that work over solutions that are cheap.

The policy comes from Amazon's massive retail division, known internally as "Stores." It represents a formal attempt to scale AI usage across thousands of teams while keeping adoption grounded in practical outcomes rather than technological novelty.

Build Now, Optimize Costs Later

For a company famous for its frugality, one guideline stands out. Amazon is explicitly telling engineers to spend what it takes to get something working.

“We prioritise working, effective solutions over cheap ones. This means we will build now, then optimise for compute cost later.”

— Amazon's internal AI policy

This inverts the typical cost-first thinking many engineering teams default to. Amazon is betting that shipping functional AI features quickly matters more than minimizing cloud bills in the early stages.

Not Everything Needs an LLM

The guidelines push back against the instinct to apply large language models everywhere. Amazon wants engineers to pick the best tool for each problem, even if that tool is not AI at all.

“We will use the best approach to solve the problem we face. Sometimes that will require AI, and sometimes the AI will be an LLM, but not always.”

— Amazon's internal AI policy

This acknowledges what many engineers already know: LLMs are powerful but not universally appropriate. A rules-based system or a simpler ML model often does the job faster, cheaper, and more reliably.

Stay Off the Bleeding Edge

The most direct guidance warns against constantly upgrading to the newest AI models. Amazon wants teams to stick with stable, proven technologies unless the benefits of switching clearly outweigh the costs.

The policy states: "We will evaluate and retain flexibility to switch if the benefits outweigh the costs, sometimes foregoing the newest improvements." This is a direct counter to the pressure many engineering teams feel to adopt each new model release from OpenAI, Anthropic, or Google.

Domain Experts Stay in the Loop

Amazon is clear that AI teams will not become experts in every business area they touch. Instead, domain experts must participate actively in AI pilots, bringing their knowledge and time to the table.

The policy reads: "We will rely on existing teams' expertise and will not become domain experts in your area. Participating in our pilots requires bringing your domain expertise and time investment."

This sets expectations on both sides. AI teams provide tools and infrastructure. Business teams provide context and judgment. Neither can succeed alone.

Scale Beats Customization

One guideline will frustrate teams hoping for tailored solutions. Amazon explicitly says it will not accommodate every customer preference. The goal is building systems that work across hundreds of teams, not bespoke tools for each one.

"Although we will aim to delight our customers, we will not accommodate all their preferences," the policy states. This reflects the reality of operating at Amazon's scale. Customization creates maintenance burdens. Standardization enables speed.

Integration Over Add-On

Amazon spokesperson Montana MacLachlan told Business Insider that the real gains come from embedding AI throughout the development lifecycle, not treating it as a feature to bolt on at the end.

"Amazon's Stores engineering teams found that integrating AI across the full development lifecycle, not just bolting it on as an afterthought, delivers the most meaningful gains in what we're able to invent for customers and how quickly we can deliver it," MacLachlan said.

Frequently Asked Questions

What are Amazon's six AI engineering tenets?

Amazon's internal AI guidelines cover prioritizing working solutions over cheap ones, using the best tool for each problem (not always AI), avoiding bleeding-edge technologies, requiring domain expert participation, building for scale over customization, and integrating AI throughout development rather than adding it later.

Why is Amazon telling engineers to avoid bleeding-edge AI?

Amazon wants teams to use stable, proven technologies unless the benefits of switching to newer models clearly outweigh the switching costs. This prevents constant disruption from chasing each new model release.

Does Amazon use LLMs for everything?

No. Amazon's policy explicitly states that engineers should use the best approach for each problem, and that will not always be an LLM. Simpler solutions may be faster, cheaper, or more reliable.

What is Amazon's AI Native strategy?

Amazon's AI Native strategy aims to scale AI usage across thousands of internal teams while tracking adoption closely. The six engineering tenets provide a practical framework for this expansion.

Source: mint / Aman Gupta