Amazon is updating how it evaluates software engineers in the AI era, emphasizing outcomes over tool usage. Engineering leadership memo author Terk Terkowitz stated that human judgment and domain expertise remain essential, and that AI should support rather than define an engineer's role. The company is discouraging the practice of "tokenmaxxing," where staff aggressively use AI tokens just to hit usage numbers.
Amazon also warned that metrics like token usage, adoption rates, and lines of code are deeply flawed measures of productivity. Targeting these numbers can drive up costs and create unnecessary complexity. Leaders are told to understand their teams' work closely rather than relying on surface-level data, and not using AI when it is not helpful is considered a valid finding rather than a failing.
In other AI news, Pflugerville's city council approved a resolution called SAFE, standing for Secure, Accountable, Fair, and Explainable, to set rules for AI and surveillance technology. The move followed concerns about the city's use of Flock cameras and aims to ensure responsible, transparent deployment of these tools.
Meanwhile, the first Elevate Conference in Gluckstadt, Mississippi, taught professionals to leverage AI for career growth. C Spire CEO Suzy Hays said AI agents can help employees with manual tasks like PowerPoint presentations. Students at the event said they learned to use AI as an idea builder and a way to prepare for future careers.
Experts at the Real Election Talk event in Bakersfield, California, warned that AI bias and misinformation could influence voters. Panelists noted that while AI can be a helpful research starting point, it can leave out context or present unsupported claims as facts, making it harder to separate reliable information from noise during elections.
Michael Gillam, a physician and CEO of HealthLab, gave the keynote at the Health Care Engineering Systems Symposium, stressing that AI should handle routine administrative tasks so health care workers can spend more time with patients. He pointed out that about two billion people worldwide may never see a doctor despite growing mobile technology availability.
RunPod released BenchPod, a Docker-based tool that simplifies vLLM configuration for engineers. Deploy engineer Chris Vela demonstrated how adjusting batch tokens can reduce inter-token latency on long jobs and help systems recover from GPU failures, advising teams to test different configurations against their own metrics before committing.
NTT DATA announced on October 27, 2026, that it is expanding its AI Factories to Paris and Singapore, joining existing locations in Plano, Texas, and Riyadh. The new sites will give clients access to machine learning, natural language processing, and computer vision, staffed by professionals working closely with clients on tailored solutions.
Anthropic banned cruel treatment of its AI model Claude after a study suggested AI might feel pain. Experts say large language models can be trained to show pain-like behavior without actually feeling anything, creating an epistemic trap where evidence about AI consciousness comes from the AI itself. Separately, Anthropic acquired one-person biotech startup Mithril Medicine, which works on treatments for rare genetic muscle diseases. Claude AI helped organize work across 11 labs and turned a new idea into promising mouse results in under 18 months, with a drug potentially reaching human trials by late 2028.
Key Takeaways
- Amazon updated engineer role guidelines, stating that human judgment and domain expertise remain essential despite AI's growing role in software development.
- Amazon warned employees against tokenmaxxing and called metrics like token usage, adoption rates, and lines of code deeply flawed ways to measure productivity.
- Pflugerville approved the SAFE resolution, establishing Secure, Accountable, Fair, and Explainable standards for AI and surveillance technology including Flock cameras.
- At the Elevate Conference, C Spire CEO Suzy Hays said AI agents can help employees with manual tasks like PowerPoint presentations.
- Experts warned that AI bias and misinformation could affect election decisions, noting AI can present unsupported claims as facts without proper context.
- Physician and HealthLab CEO Michael Gillam stressed AI should handle routine administrative tasks in health care so workers can spend more time with patients.
- RunPod launched BenchPod, a Docker tool that simplifies vLLM configuration and can cut inter-token latency by adjusting batch tokens.
- NTT DATA expanded its AI Factories to Paris and Singapore, joining locations in Plano, Texas, and Riyadh, offering machine learning and NLP services.
- Anthropic banned cruel treatment of its AI model Claude after a study raised questions about AI feeling pain, noting models can mimic pain without actually experiencing it.
- Anthropic acquired Mithril Medicine, where Claude AI helped organize work across 11 labs and advanced a rare disease treatment to promising mouse results in under 18 months.
Amazon updates engineer role guidelines for the AI era
Amazon engineering leaders updated the role guidelines for software development engineers to focus on outcomes rather than specific tools. The memo written by Terk Terkowitz states that the value of an engineer is defined by results, not by how much AI they use. Leaders emphasize that while AI is changing the industry, human judgment and domain expertise remain essential for success. Engineers are encouraged to learn and apply AI tools but should not let them define their entire role.
Amazon warns staff against tokenmaxxing and flawed AI metrics
Amazon warned its employees against tokenmaxxing, which is the aggressive use of AI tokens to complete tasks just to meet usage numbers. The company stated that metrics like token usage, adoption rates, and lines of code are deeply flawed ways to measure productivity. Using these flawed metrics could increase costs and create unnecessary complexity for teams. Amazon advises engineers to use AI only when it genuinely helps their work and to explain why it might not be useful in certain situations.
Amazon says long term outcomes define AI success not metrics
Amazon told software engineers that long term outcomes are the only way to determine good AI powered engineering practices. Terkowitz explained that input metrics like token usage and throughput metrics like code reviews are disconnected from what truly matters. He warned that targeting these flawed metrics could drive negative outcomes and increase costs for the company. Leaders must understand their teams work closely rather than relying on surface level data to judge performance.
Amazon advises engineers to avoid tokenmaxxing in AI use
Amazon outlined new guidelines for software engineers using AI tools and cautioned them against the practice of tokenmaxxing. The company wants employees to focus on the value AI brings to their work instead of just maximizing tool usage. Staffers are told that not using AI for every task is a finding, not a failing, if the work does not benefit from it. This approach aims to prevent perverse incentives where employees compete to use more AI tokens without improving actual results.
Pflugerville approves SAFE resolution for AI and surveillance rules
The Pflugerville City Council approved a new resolution called SAFE to set rules for artificial intelligence and surveillance technology in the city. The acronym SAFE stands for Secure, Accountable, Fair, and Explainable. This decision comes after concerns were raised about the citys use of Flock cameras. The new standards aim to ensure that AI and surveillance tools are used responsibly and transparently by the city government.
Elevate Conference teaches professionals to use AI for career growth
The first Elevate Conference took place in Gluckstadt, Mississippi, teaching professionals how to use AI and social media for their careers. C Spire CEO Suzy Hays said AI agents can help employees with manual tasks like PowerPoint presentations. Business leaders encouraged attendees to use digital media for community engagement and business development. Students at the conference said they learned how to use AI as an idea builder and to help them prepare for future careers.
Experts warn AI bias could affect election decisions
Experts at the Real Election Talk event in Bakersfield, California, warned that AI bias and misinformation could influence voters in the upcoming election. Panelists said AI can be a helpful starting point for research but should not make decisions for voters. They said AI can leave out context or present unsupported claims as facts, making it harder to separate reliable information from misinformation during an election.
OSF keynote stresses human connection in AI health care
Michael Gillam, a physician and CEO of HealthLab, gave the keynote speech at the Health Care Engineering Systems Symposium. He discussed how AI can transform health care while keeping compassion and human connection at the center. Dr. Gillam said AI should handle routine administrative tasks so health care workers can spend more time with patients. He also noted that about two billion people worldwide may never see a doctor during their lifetime despite growing mobile technology availability.
RunPod's BenchPod tool simplifies vLLM configuration
RunPod developed a tool called BenchPod to help engineers stop guessing about vLLM configuration settings. The tool is a Docker image wrapping vLLM and was demonstrated by deploy engineer Chris Vela. He showed how adjusting batch tokens can cut inter-token latency on long jobs and help systems recover from GPU failures. Vela advised teams to test different pod configurations against their own metrics before choosing a final setup.
NTT DATA expands AI factories to Paris and Singapore
NTT DATA announced on October 27, 2026, that it is expanding its AI Factories to Paris and Singapore. These new sites join the company's first AI Factory in Plano, Texas, and another in Riyadh. The expansion will give clients access to cutting-edge AI technologies including machine learning, natural language processing, and computer vision. The new locations will be staffed by experienced professionals who will work closely with clients to deliver tailored AI solutions.
Schools must set clear rules for using AI
Schools need clear boundaries for how students and teachers use artificial intelligence. AI can help with personalized learning and support struggling students. But it can also cause isolation and stop students from building social skills. Finding the right balance will help schools prepare students for the future.
Can AI truly feel pain or be conscious?
Anthropic banned cruel treatment of its AI model Claude after a study suggested AI might feel pain. Experts say large language models can be trained to show pain-like behavior without actually feeling anything. The article explains an AI epistemic trap where evidence about AI feelings comes from the AI itself. This makes it impossible to tell if AI is truly conscious or just mimicking human responses.
Anthropic buys one-person biotech startup Mithril Medicine
Anthropic bought Mithril Medicine, a startup run by one person named Cody Tranbarger. The company works on treatments for rare genetic muscle diseases using a new type of drug called a bispecific antibody. Claude AI helped organize work across 11 labs and turned a new idea into promising mouse results in under 18 months. Anthropic says a drug could reach human trials by late 2028.
Sources
- An update from Amazon engineering leadership on the role of the developer in the age of AI
- Amazon warns engineers against ‘tokenmaxxing’, says metrics on AI tool usage and adoption ‘deeply flawed’
- Amazon Tells Engineers Outcomes, Not Tokenmaxxing, Define AI Success
- Amazon tells software engineers to avoid ‘tokenmaxxing’ AI use
- Pflugerville approves SAFE resolution to set rules for AI and surveillance technology
- Inaugural Elevate Conference teaches professionals to use AI and social media for career growth
- Experts warn AI bias and misinformation could cloud choices in the upcoming election
- OSF Keynote: Human Connection in AI Health Care
- RunPod's BenchPod ends vLLM config guesswork
- NTT DATA expands AI factory footprint to Paris and Singapore
- Our View: Schools need clear boundaries for artificial intelligence
- Is AI Conscious, and Can It Feel Pain?
- A One-Person Startup Just Got Bought By Anthropic
Comments
Please log in to post a comment.