How I use AI (Oct 2026)
A Salesforce AI Research engineer offers a captivating 2026 retrospective on how AI has profoundly reshaped daily life and work. From agents writing 90% of his code to the contentious rise of AI art, the piece highlights both miraculous productivity gains and troubling ethical dilemmas. It's a stark, near-future dispatch that fuels Hacker News's ongoing debate about AI's transformative power and its cost to human craft and authenticity.
The Lowdown
In a follow-up to his 2025 reflections, a research engineer (now at Salesforce AI Research) details his AI usage in October 2026, revealing a radically transformed landscape. He spends around $3,000/month on tokens and has even drafted a personal AI policy to navigate the pervasive influence of these tools.
- Coding: The author declares "Writing code is solved." He now manages 5-6 coding agents for everything from prototyping to PRs, enjoying unprecedented speed and reduced drudgery. However, this comes with cognitive fatigue, a sense of lost craft, and a crucial ongoing need for human architectural oversight and taste. Agentic PRs and code reviews are becoming standard.
- Research & Search: Generative AI now dominates search, with Google's AI summaries as the default. LLMs excel as patient teachers for established concepts and are surprisingly effective for literature searches, bypassing clunky traditional tools. Yet, product search remains poor, and AI is still unreliable for collaborative novel research, often leading to an unproductive "AI fog."
- Summarization, Transcription & Voice: These tasks are
The Gossip
Code's Conquered or Contested?
The author's bold claim that "writing code is solved" ignited a fierce debate. Many commenters strongly disagreed, pointing out that AI-generated code often requires extensive human oversight, leading to "vibe-code" or "slop." Critics emphasized the continued necessity for human taste, architectural design, and bug diagnosis, noting that AI frequently bakes in incorrect assumptions. Conversely, some conceded that AI excels at boilerplate, low-quality code improvements, and rapid prototyping, especially in specific language contexts like Go. The discussion also touched on the evolving definition of 'coding'—from writing lines to managing agents—and whether training data degradation would eventually curb AI's coding prowess.
Token Tally and Tech Titans
A point of contention arose from the author's stated $3,000/month token spend, which seemed incongruous with his $27/month personal subscriptions. The author clarified that his company covers the bulk of these work-related token costs, highlighting the enterprise-scale investment in advanced AI usage. This led to discussions about the economics of AI, with some suggesting that for such heavy usage, investing $36k-$44k in powerful local GPU hardware to run models 24/7 might be more cost-effective than continuous cloud inference fees.
Aesthetic Arguments and AI Artistry
The author's nuanced take on AI art, disliking AI image/video 'slop' but enjoying non-commercial AI music, sparked passionate discussion. Proponents cited examples of creative AI-generated short films and music, arguing that AI is merely another tool for human expression and that art is inherently subjective. They also suggested AI music can offer novelty superior to mass-produced pop. Detractors vehemently argued that AI art lacks human soul, expression, and meaning, dismissing it as hollow 'slop' primarily used for cynical commercial exploitation. The debate frequently circled back to the fundamental definition of art and whether a machine can truly create it without consciousness or human experience.
Human Hazards and Helper Hypotheses
Commenters largely resonated with the author's experience of 'fatigue' from managing multiple AI agents, describing it as 'overstimulating & exhausting' in ways traditional coding wasn't. Many lamented the 'loss of craft' and the meditative joy of hands-on programming, questioning whether perceived productivity gains were worth sacrificing professional purpose. Some also debated whether an AI research engineer's advanced usage could truly be considered 'typical.' The broader impact on the future of work was also touched upon, with concerns raised about AI's potential to automate tech-adjacent roles like sysadmin tasks.