The editorial argues that AGPL-3's SaaS loophole protection was load-bearing against a threat that never materialized. None of Cal.com's actual competitors (SavvyCal, Motion, Reclaim, Akiflow, Calendly) forked the codebase — they built their own schedulers because scheduling is a two-week MVP, not defensible IP. Blaming AI scraping obscures that copyleft never protected Cal.com from its real competition.
The editorial contends Cal.com is a YC-backed Series-B company whose investors want a wider moat, and AI is a more sympathetic villain than admitting the AGPL constraints no longer serve the cap table. The Hacker News thread's top comments converge on this reading, treating Richelsen's AI framing as post-hoc justification for a commercial pivot.
The Strix.ai post argues Cal.com's relicense is being misread as evidence that AI has broken open source, when the real lesson is that Cal.com misdiagnosed its own competitive dynamics. Open source models still work; Cal.com just picked a license targeted at a threat (hosted forks) that was never the actual risk in its market.
Richelsen frames the relicense as a response to AI scrapers ingesting the codebase and LLMs regurgitating Cal.com's implementation into competitors faster than AGPL compliance can be enforced. In this view, copyleft's network-service clause is structurally mismatched to a world where code flows through training data rather than direct forks.
Cal.com, the Y Combinator-backed scheduling tool that built its brand on being the AGPL-licensed alternative to Calendly, announced it is relicensing its core repository away from AGPL-3.0. CEO Peer Richelsen framed the decision around AI: scrapers ingesting the codebase, LLMs regurgitating Cal.com's implementation into competitors, and the company's inability to enforce AGPL compliance at the speed code now moves.
The Hacker News thread — 321 points and climbing — is not buying it. The top comments zero in on a simpler explanation: Cal.com is a Series-B SaaS company that no longer wants the constraints it accepted when it needed the community, and AI scraping is a more palatable villain than "our investors want a wider moat." The Strix.ai rebuttal, which prompted the discussion, makes the case bluntly: open source is not dead, but Cal.com just learned the wrong lesson from its own data.
The wrong lesson is the headline claim that AI broke the AGPL's protections. It didn't. AGPL was already failing at its stated purpose long before any transformer model existed, and Cal.com's own competitive landscape is the proof.
Start with what AGPL-3 actually does. It closes the "SaaS loophole" in GPL: if you modify AGPL code and expose it over a network, you must publish your modifications. That is it. It is a copyleft clause targeted at a specific commercial pattern — a competitor forking your code, running it as a hosted service, and keeping their changes private.
Count the competitors who did that to Cal.com. The answer is roughly zero. SavvyCal, Motion, Reclaim, Akiflow, Calendly itself — none of them forked Cal.com. They wrote their own schedulers, because scheduling is a two-week MVP for a competent team, not a defensible piece of IP. The AGPL clause was load-bearing against a threat that never materialized, because the threat model was wrong. Scheduling software is not a database engine or a search index; there is no deep technical moat for a license to protect.
This is the generalizable failure, and it applies to a long list of "open core" companies now quietly re-papering their licenses. MongoDB went SSPL in 2018. Elastic went SSPL/Elastic License in 2021. Redis went RSALv2/SSPL in 2024. HashiCorp went BSL in 2023. Each time, the stated villain was a hyperscaler or a cloud reseller. Each time, the underlying problem was the same: a permissive-adjacent license is a legal instrument, not a commercial one, and no license survives contact with a competitor who has enough capital to pay engineers to retype your ideas.
Now layer AI on top. Cal.com's new claim is that LLMs accelerate the retyping — a junior engineer with Cursor can reproduce a scheduling flow in an afternoon. This is true, and it is also true of every SaaS codebase in every license, public or private. Closed-source will not save you from this; it will just slow down the copy by the time it takes to reverse-engineer the API surface, which for a scheduling product is measured in days. The AI argument is real as a trend and hollow as a license justification. It is being used because "we want to recapture the margin that our permissive license gave away" polls badly on Hacker News.
The community reaction reflects this. A recurring comment pattern in the HN thread: users who deployed self-hosted Cal.com specifically because of AGPL are now evaluating Cal.com forks, Rallly, and writing their own calendar endpoints against Google/Microsoft Graph. The same license clause that failed to block commercial competitors succeeded spectacularly at attracting the self-hosters who now feel betrayed, which is the exact inverse of the business outcome Cal.com wanted.
If you ship open source, stop treating the license as a moat. Pick the license that matches the behavior you actually want from your users, not the behavior you want to prevent from hypothetical competitors. If you want maximum adoption, go MIT or Apache-2 and charge for hosting, support, or a proprietary control plane. If you want contribution reciprocity from modifiers, go AGPL but do not expect it to protect revenue. If you want to prevent hyperscaler resale, you need BSL or a commercial license — and you need to be honest with your community about that from day one, not four years in.
If you depend on an AGPL-licensed tool in production, assume the license can change. Build your dependency graph with a 90-day exit plan: which fork would you move to, who maintains it, how hard is the migration? Cal.com is self-hostable on Docker with a Postgres backend; the forks will exist within weeks. The harder case is upstream libraries buried in your build — this is where license audits pay for themselves.
If you are building a new open-source commercial product in 2026, the math has shifted. Permissive licenses plus a managed service plus genuine operational excellence (migrations, backups, observability, SSO) is now a more defensible strategy than copyleft plus "please comply." The AGPL-era bet was that legal friction would create business friction for competitors. That bet did not pay off. The new bet is that ops excellence and distribution create friction competitors can't paper over with a rewrite.
Expect two or three more mid-stage open-source SaaS companies to announce similar relicenses in 2026, with AI cited as the proximate cause each time. The underlying driver will be the same VC math that pushed MongoDB and Elastic: growth expectations that require margin expansion, and a recognition that the original license gave away leverage the company now needs back. The honest version of the Cal.com announcement would have read: we misjudged what a license could protect, we are correcting that, and we are sorry to the self-hosters. The AI framing is the corporate-comms version of the same admission. Either way, the lesson for the rest of the industry is the one the license community has been arguing about since SSPL: ship the business model first, pick the license to match it, and do not expect the copyleft clause to do work that only distribution, brand, and product velocity can do.
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.