Industry experts classify operational delays into intent friction, which covers repetitive execution tasks, and problem friction, which arises when the underlying challenge remains ambiguous. While large language models excel at removing intent friction, they struggle with ambiguity and frequently produce standardized or sycophantic results. Direct engagement with actual users before proposing solutions exposes flawed assumptions and forces teams to rethink strategic priorities. Teams at thoughtbot deploy structured shaping sprints to deliberately engage this ambiguity before committing engineering resources to development.
Key points
Eliminating all friction from product workflows increases the risk of delivering tools th…: Eliminating all friction from product workflows increases the risk of delivering tools that fail to address real user needs.
Protecting problem friction ensures teams actively debate different opinions, critique ea…: Protecting problem friction ensures teams actively debate different opinions, critique early designs, and uncover critical product risks.
Conducting structured shaping sprints allows organizations to test unverified ideas befor…: Conducting structured shaping sprints allows organizations to test unverified ideas before devoting costly development resources to them.
Sources checked by our analysis pipeline: 1
Today, with the rise of LLMs that can generate a product brief or full prototypes in minutes, friction between a product concept and a polished deliverable is getting smaller and smaller. The tools and processes for product discovery and shaping have changed rapidly. Founders and product teams are shaping ideas with AI and can move quickly but this loss of friction can lead to shipping the wrong thing if we’re not careful.
Categorizing friction
The debate around the question “is the way AI removes friction a good thing” is certainly not new but I recently heard Fabio Cortez speak at Webflow Conf and this quote stuck with me:
there are two kinds of friction, one we want to automate, the friction between me and my intent, the repetition, translation, reconstruction happening between stakeholders, we should automate that. But the friction between us and the problem we’re trying to solve, we should not automate, we should protect it always, that’s where the magic happens, the ambiguity, the critique, the exploration, that’s the friction to protect.
Recent evaluations establish the model firmly near the pinnacle of the frontier AI landscape, sitting just behind two market-leading offerings. These performance metrics suggest high-end development tasks can now be achieved with far more economical resource expenditures. The data reflects a shifting competitive balance where leading-tier capabilities are becoming viable at significantly reduced price points.
The reported offshore data infrastructure highlights how advanced hardware can flow across borders despite geopolitical scrutiny. Working through intermediary entities allows computing resources to reach facilities abroad, raising international security concerns regarding defense-adjacent research programs. This mechanism demonstrates the operational strategies employed to secure access to critical foreign technology.
Enterprises seeking tailored artificial intelligence can now comprehensively update model parameters without provisioning clusters or managing distributed training environments. This serverless capability allows smaller specialized models to match or exceed frontier systems, lowering per-token inference costs across millions of queries. By updating weights uniformly rather than relying on low-rank approximations, systems can fully absorb domain-specific knowledge at scale.