A personal account published this week on hackernews details one developer's deliberate decision to stop using Anthropic's Claude and migrate their daily AI assistant work to Pi, the conversational AI product from Inflection AI. The post, written by Caleb on his personal software blog at caleb.software, describes the transition as driven primarily by a combination of subscription cost concerns and inconsistent service behavior from Claude over time.
The author does not publish specific dollar figures for what he was spending on Claude's subscription tier, but frames the cost-to-value ratio as having degraded sufficiently to prompt a full switch rather than a reduction in usage. Pi, by contrast, is currently available without a paid subscription wall for its core conversational features, which the author identifies as a meaningful practical difference. The piece generated notable discussion on hackernews, where readers debated whether Pi's conversational quality holds up for technical and coding tasks — the specific workload Caleb describes relying on most heavily.
The post does not cite performance benchmarks or third-party evaluations, making it an anecdotal account rather than a systematic comparison. Inflection AI, which originally launched Pi in 2023 and subsequently restructured after a significant portion of its leadership and staff moved to Microsoft in 2024, has continued operating the Pi product independently. Anthropic, the maker of Claude, raised $7.3 billion in funding from Amazon and other investors through 2024 and 2025 and has continued releasing updated Claude model versions, most recently in its Claude 3.x and Claude 4 series.
What the general tech press won't surface here is the dependency risk embedded in any household or small operation that has integrated a single commercial AI assistant deeply into communications, planning, or information workflows. When a paid AI service changes its pricing structure, degrades in quality, or experiences outages — all of which Claude users have reported in community forums at various points in 2025 and 2026 — there is no analog fallback built into most people's routines the way there might be for, say, a power outage. The developer in this post had the technical fluency to evaluate an alternative and migrate. Most users don't perform that audit until the disruption has already cost them time or money, and they haven't maintained parallel familiarity with any substitute tool. Diversity across AI platforms, treated the same way resilient households treat diversity in communication channels or food sourcing, is an underexamined dimension of practical digital continuity.
