I coding has emerged as one of the most lucrative and productive use cases for the tech among enterprises. But businesses may be jumping the gun.
Executives at companies like Spotify and Canva have admitted that their developers hardly write code anymore, leading many to shift their responsibilities to orchestration, rather than development. And while the jury is still out on whether AI can entirely automate this work, enterprises are using their AI shifts to justify substantial layoffs.
The numbers are starting to pile up, with companies like Atlassian, Oracle, Block and Amazon slashing thousands, leaving large swaths of software engineers without work.
Still, even though enterprises seem ready for AI to automate their coders, AI might not be ready for the big leagues just yet. It’s a lesson that Amazon seems to be learning in real time: After a string of outages connected to AI-generated code cost the company millions of lost orders, the e-commerce giant has ordered a 90-day safety guardrail on code written for “tier-1 systems,” or those that directly face consumers, according to Business Insider.
Outages linked to the use of AI coding tools, including the company’s own assistant Q, have emerged since the third quarter of last year, Business Insider reported. The latest outage occurred in early March, taking out the company’s website and shopping app for several hours.
As a result, developers are going to be on a tighter leash:
- Amazon’s developers are now required to document changes more stringently and get additional approvals from two other people on staff before making changes to code.
- The company is also developing safeguards that combine agentic and deterministic, or rules-based, guardrails.
If there’s one thing we know to be true about AI, it’s that these models are not infallible. Though these tools might be tempting for the speed that they unlock, speed can often be the enemy of quality, changing the “risk profile” that enterprises have to consider, Satyam Dhar, staff software engineer at Galileo and former Amazon and Adobe engineer, told The Deep View. AI tools can create a lot of code that looks like it’s ready for production before it’s actually been pressure tested, he said. “That’s where things get dangerous.”
“The discipline doesn’t change,” said Dhar. “AI helps with scaffolding, but the engineering judgment still has to come from humans.”
Our Deeper View
Though Amazon’s slip-ups may serve as a lesson for other enterprises to take it slow, the momentum around coding assistants may make it difficult to heed. Along with automating coding itself, major AI firms like OpenAI and Anthropic are now throwing AI at code review and security processes. Enterprises, faced with both the temptation to supercharge productivity and the pressure to keep up with competitors, may not bear in mind the risks that they’re taking on by reaching for the shiny new toy.