Tech
‘Tokenmaxxing’ is making developers less productive than they think
There’s an old saw in management: What you measure matters. And, typically, you get more of whatever you’re measuring.
Software engineers have debated productivity metrics for decades, starting with lines of code. But as the new generation of AI coding agents delivers more code than ever, what their managers ought to be measuring is less clear.
Enormous token budgets — essentially, the amount of AI processing power a developer is authorized to consume — have become a badge of honor among Silicon Valley developers, but that’s a very weird way to think about productivity. Measuring an input to the process makes little sense when you presumably care more about the output. It might make sense if you’re trying to encourage more AI adoption (or selling tokens), but not if you’re trying to become more efficient.
Consider the evidence from a new class of companies operating in the “developer productivity insight” space. They’re finding that developers using tools like Claude Code, Cursor, and Codex generate a lot more accepted code than they did before. But they also find that engineers have to return to revise that accepted code far more often than before, undercutting claims of increased productivity.
Alex Circei, the CEO and founder of Waydev, is building an intelligence layer to track these dynamics; his firm works with 50 different customers that employ more than 10,000 software engineers. (Circei has contributed to TechCrunch in the past, but this reporter had never met him before.)
He says that engineering managers are seeing code acceptance rates of 80% to 90% — meaning the share of AI-generated code that developers approve and keep — but they’re missing the churn that happens when engineers have to revise that code in the following weeks, which drives the real-world acceptance rate down between 10% and 30% of generated code.
The rise of AI coding tools led Waydev, founded in 2017 to provide developer analytics, to totally rework its platform in the last six months to address the proliferation of rapid coding tools. Now, the company is releasing new tools that track the metadata generated by AI agents, offering analytics on the quality and cost of their code to provide engineering managers with more insight into both AI adoption and efficacy.
Techcrunch event
San Francisco, CA
|
October 13-15, 2026
While analytics companies have an incentive to highlight the problems they find, the evidence is mounting that large organizations are still figuring out how to use AI tools efficiently. Major companies are noticing — Atlassian acquired DX, another engineering intelligence startup, for $1 billion last year, to help its customers understand the return on investment on coding agents.
The data from across the industry tells a consistent story: More code is being written, but a disproportionate amount of it isn’t sticking.
GitClear, another company in this space, published a report in January that found AI tools increased productivity, but also that its data showed “regular AI users averaged 9.4x higher code churn than their non-AI counterparts” — more than double the productivity gains the tools provided.
Faros AI, an engineering analytics platform, drew on two years of customer data for its March 2026 report. The finding: code churn — lines of code deleted versus lines added — had increased 861% under high AI adoption.
Jellyfish, which bills itself as an intelligence platform for AI-integrated engineering, collected data on 7,548 engineers in the first quarter of 2026. The firm found that the engineers with the largest token budgets produced the most pull requests (proposed changes to a shared codebase), but the productivity improvement didn’t scale. They achieved two times the throughput at 10 times the cost of tokens. In other words, the tools are generating volume, not value.
These kinds of statistics ring true when you talk to developers, who are finding that code review and technical debt are stacking up, even as they revel in the freedom of the new tools. One common finding is the difference between senior and junior engineers, with the latter accepting far more AI-generated code, and dealing with a larger amount of rewriting as a consequence.
Still, even as developers work to understand exactly what their agents are up to, they don’t anticipate turning back anytime soon.
“This is a new era of software development, and you have to adapt, and you are forced to adapt as a company,” Circei told TechCrunch. “It’s not like it will be a cycle that will pass.”
Tech
Netflix plans to add a vertical video feed, use AI for recommendations
Netflix is going to launch a TikTok-like vertical video feed within its apps this month, and plans to use AI broadly for content creation and recommendations, the company said on Thursday.
Netflix has been testing a vertical video feed since last year. The short video feature could aid users with discovering video podcasts, along with the current slate of shows and movies. The company is also leaning more into using AI for recommendations after launching a ChatGPT-powered search feature last year.
“We have been in personalization and recommendation for two decades, but we still see tremendous room to make it better by leveraging newer technologies,” Netflix co-CEO Gregory Peters said during the company’s first-quarter conference call. “Recommendation systems based on new model architectures not only improve current personalization but also let us iterate and improve more quickly — adding support for different content types much more efficiently.”
Co-CEO Ted Sarandos said he sees AI tools improving the entire content creation process. “In general, we expect GenAI to make content better; better tools, better processes […] It takes a great artist to make great art, and AI won’t change that. But AI will give those artists better tools to bring those visions to life,” he said.
Last month, Netflix bought Ben Affleck’s AI creation company InterPositive, which, Sarandos said, has garnered interest from creators.
“With our acquisition of InterPositive, we think it accelerates our GenAI capability because it is proprietary technology created specifically for filmmakers and filmmaking, different from other GenAI video applications. While our ownership of InterPositive is very new, we have generated interest with creators who have spent time with the tools, and we are seeing momentum build around adoption,” he noted.
Netflix also mentioned that it wants to use AI to improve its ad suite, and allow for new formats and customization to get better returns. The company expects to generate ad revenue of $3 billion this year.
Techcrunch event
San Francisco, CA
|
October 13-15, 2026
Netflix reported revenue of $12.25 billion in Q1 2026, up 16.2% year-year-year, and said profit jumped 83% to $5.28 billion. Alongside the first-quarter results, Netflix said its co-founder and chair, Reed Hastings, is leaving the company’s board this summer.
Notably, the company hiked subscription prices in the U.S. late last month, which could have a positive impact next quarter. The company said it ended 2025 with 325 million paying subscribers.
Tech
Bluesky confirms DDoS attack is cause of continued app outages
Bluesky’s website and app are still struggling on Friday after experiencing service interruptions that chief operating officer Rose Wang attributed to an ongoing cyberattack.
On Thursday evening, the social media company confirmed that a “sophisticated Distributed Denial-of-Service (DDoS) attack” was to blame for the issues, which had originally started on April 15 at around 8:40 p.m. ET.
Distributed denial-of-service attacks often involve pummeling apps or websites with large amounts of junk web traffic aimed at overloading and knocking its servers offline. While these kinds of cyberattacks do not involve intrusions into a company’s systems, these incidents can still be disruptive to both the company and its users.
In a post on the Bluesky account, the company shared the cause of the problem and noted that the attack was “impacting our operations, with users experiencing intermittent interruptions in service for their feeds, notifications, threads, and search.”
Bluesky said that it has not seen any evidence of unauthorized access to private data, however.
When originally reached for comment on Thursday, Bluesky only pointed us to the status.bsky.app page and account (@status.bsky.app) for updates. The company did not provide an estimated time for a fix.
The network’s status page is currently not working, however.
Bluesky said it will provide another update on the status of the attack and its mitigation by 1 p.m. ET on Friday.

Because the outages are intermittent, the Bluesky site and app will load at times, slowly, and other times will display error messages.
For instance, switching to a particular feed within the app could display a message that says, “This feed is currently receiving high traffic and is temporarily unavailable. Please try again later. Message from server: Rate Limit Exceeded.”

Popular feeds like Discover or the official Bluesky Team’s feed often see this problem, even as users’ own personal feeds are functional.
Other times, like when trying to visit a user’s profile, the site will display an error message, forcing you to refresh and try again.

Bluesky protocol engineer Bryan Newbold remarked around 3:46 a.m. ET on Wednesday, “oof, our services are getting hit pretty hard tonight.”
Notably, the service disruptions are impacting Bluesky, but other communities, like Blacksky, that run their own infrastructure on the underlying protocol that powers the decentralized social network, are still functioning.
Blacksky’s team told TechCrunch that the Bluesky outage has led to a “significant spike” in migration requests from Bluesky users over the past 12 hours, as users, devs, and other ATmosphere founders like Sebastian at Eurosky have been promoting its services.

It was clear that Bluesky’s team was in a hectic state this week while facing these issues, as one message on its status page had a typo: ” investigating an incident with service in one of our reginos [sic].”

Tech
With US spy laws set to expire, lawmakers are split over protecting Americans from warrantless surveillance
A long-running law that has allowed U.S. intelligence agencies to collect and analyze huge amounts of overseas communications without needing search warrants is set to expire next week, and lawmakers are in a deadlock over whether to allow the Trump administration to extend it without any changes.
Known as Section 702 of the Foreign Intelligence Surveillance Act (FISA), the law allows the National Security Agency, the CIA, the FBI and other federal intelligence agencies to record overseas communications that flow through the United States without needing individualized search warrants.
In sweeping up much of the world’s communications, the agencies also collect unfathomable amounts of information, including phone logs and emails, on Americans who interact with people subject to surveillance overseas. This data is collected despite constitutional protections that should shield Americans and people in the United States from government surveillance.
But ahead of the law’s expiry on April 20, a bipartisan, pro-privacy group of House lawmakers and Senators are calling for sweeping changes to FISA, arguing that the changes are “essential” for protecting the privacy rights of Americans.
Some lawmakers are calling for widespread reforms following years of scandals and surveillance abuses across successive U.S. administrations, while others are holding their vote to further their own political goals by attaching the provisions to other legislation.
A social media post from President Trump suggests that, as of this week, the White House is keen on the idea of passing a simple re-authorization without any changes to the law.
In the middle of the night into Friday, House Republicans approved to extend FISA until April 30 as a stopgap to make more time to negotiate. The Senate, set to reconvene on Monday, would still need to approve the bill by majority vote for it to pass the short-term extension.
The bipartisan group’s legislative fix is the Government Surveillance Reform Act, introduced into Congress in March by Sens. Ron Wyden (D-OR), Mike Lee (R-UT) and others, which aims to curtail some of the government’s warrantless surveillance programs. Among other things, the lawmakers seek provisions to prevent government agencies from using a “backdoor search” loophole that allows them to trawl the communications of Americans without first obtaining a search warrant.
Another key provision would prevent federal agencies from buying commercially available data about Americans from data brokers — a practice the U.S. government has long asserted that it does not need a court’s permission for.
App developers collect reams of location data from people who use smartphone apps, and then sell that information to brokers, who in turn sell that data to governments and militaries. FBI director Kash Patel confirmed in a congressional hearing in March that the FBI buys Americans’ location data without seeking court authorization.
Both Republicans and Democrats are reportedly keen on closing this loophole, which allows spy agencies to buy commercial data and use AI models to analyze billions of location points. This is currently also a sticking point in the U.S. government’s negotiations with Anthropic and OpenAI over the unrestricted use of their tools.
The American Civil Liberties Union, Electronic Privacy Information Center, and the Project on Government Oversight are among some of the privacy groups supporting the bipartisan bill.
It’s currently unclear if the bill will pass, but lawmakers say legislative reforms are necessary, especially as technological advancements make it easier for tech companies and governments to surveil people than ever before.
Wyden, the longest-serving lawmaker to sit on any congressional intelligence committee and a known privacy hawk, has warned that many lawmakers are not fully aware that multiple U.S. administrations have long relied on a secret, legal interpretation of Section 702 that “directly affects the privacy rights of Americans.” Wyden said the matter remains secret, but urged the government to declassify the information so lawmakers can discuss it.
In a post on X on Thursday, Rep. Thomas Massie (R-KY, 4th) said he would vote against the reauthorization of Section 702, after he echoed Wyden by raising concerns about how the FBI was interpreting the law.
Even if Section 702 expires on Monday, it doesn’t mark the immediate end of the U.S. government’s surveillance powers.
While lawmakers in the U.S. House of Representatives are yet to reach consensus on renewing or altering Section 702, a legal quirk would allow U.S. surveillance to continue until March 2027 unless Congress actively intervenes — even if the law expires.
This is because the secretive Washington, D.C. court that oversees the government’s compliance with FISA, known as the Foreign Intelligence Surveillance Court (FISC), asks the government each year to certify that its practices are lawful. That rubber-stamp allows the government to collect phone calls and emails for a duration of 12 months, effectively guaranteeing that the surveillance programs that rely on FISA’s legal powers will continue for at least a year.
The U.S. government also has other surveillance powers that aren’t overseen by Congress, such as Executive Order 12333, an entirely secret presidential directive that dictates most of the U.S. government’s surveillance outside of the United States. It also ensnares an unknown quantity of Americans’ private communications.
