The framing is wrong. Not even Tim Peters says it was a cancellation attempt or anything like that. Bans, especially temporary ones, are not cancellations.
To be honest though, I think cancellation is a buzzword.
It's work of art and a work of science, I'd say. Programming language dev needs both, and I think Java has learned how to do both, at least compared to its messy early years.
On the note of science, I'll credit Java's maintainers for treating their language as a technology and not merely as a product. Many languages respect the real-world physics of computers to stay relevant. Few respect the "metaphysics" until weird bugs keep emerging in the compiler and ecosystem. Bugs that can't be solved with more building, engineering, or even design. Because those bugs are not defects; they come from the concepts & axioms of the language itself, not its implementation.
Yes, feature flags are conflated with remote configs (or its more useful variety: "dynamic configs"). The difference is subtle, hence why people are talking past each other.
Feature flags are gates for whether a piece of code runs; basically, an if-condition. Remote configs are a mechanism for changing runtime values without redeploying[1].
For example:
# Feature flag — variant gate for rollout
flag = sdk.check_gate(user, "checkout_flow")
if flag == 'open':
render_new_checkout()
elif flag == 'warning':
render_warning_checkout()
else:
render_old_checkout()
# Raw remote config pulled — structured values for tuning behavior
config = sdk.get_config(user, "checkout_settings") # if the config changes based on user or context, this "remote" config is considered "dynamic"
timeout_ms = config.get("timeout_ms", 5000)
max_items = config.get("max_items", 50)
allowed_tlds = config.get("allowed_tlds", [".com", ".org"])
In practice, feature flags are implemented on top of dynamic configs[2] to manage the temporary lifecycle of a feature — aka, ship a new block of code, ramp its execution up to 100%, then delete the flag. Whereas dynamic configs are a deeper primitive meant for semi-permanent/safer operations like tuning rate limits or changing text copy on a marketing website.
As I've seen it: the forcing function that separates the concepts are experimentation platforms: when human-control of feature flags is shared (via dynamic configs) with automated & randomized assignments. That's how Statsig built their system and, in part, why they could sell for a billion. Whereas companies that ignored the difference, like LaunchDarkly, struggled outside of feature flags.
I think feature flags, remote configs, and experiments are all the same thing. Semantically they differ in how you're applying the config and interpreting the outcomes.
Humane, as a secular and universally applied sentiment, is a bit of a modern idea, once backed by common goods; shared institutions, third places, extended families, good economy, religion, etc. With those common goods fading, I see people more and more lashing out against each other; particularly in a frictionless environ that incites/outlets fantasy desires. The war of all against all from the safety of our screens; at least for the growing numbers who live their lives on the upper-case Internet.
Lower-case internet is ok as a tool for making spaces. But I reckon humane-ness, or really, virtue, is a habit built from within. And the habits the Internet rewards are generally the wrong ones.
One small edit: many also look to the Internet to meet their needs, beyond just fantasy or desire. Their lashing out comes after the disappointment: when the internet simply cannot fill the hole in them the way the common goods once did.
Statsig's core value is their experimentation platform— the automation of Data Science.
Big Tech teams want to ship features fast, but measuring impact is messy. It usually requires experiments and traditionally every experiment needed one Data Scientist (DS) to ensure statistical validity, i.e., "can we trust these numbers?". Ensuring validity means DS has to perform multiple repetitive but specialized tasks throughout the experiment process: debugging bad experiment setups, navigating legacy infra, generating & emailing graphs, compensating for errors and biases in post-analysis, etc. It's a slog for folks involved. Even then, cases still arise where Team A reports wonderful results & ships their feature while unknowingly tanking Team B's revenue— a situation discovered only months later when a DS is tasked to trace the cause.
Experimentation platforms like Statsig exist to lower the high cost of experimenting. To show a feature's potential impact before shipping, while reducing frustrations along the way. Most platforms will eliminate common statistical errors or issues at each stage of the experiment process, with appropriate controls for each user role. Engs setup experiments via SDK/UI with nudges and warnings for misconfigurations. DS can focus on higher-value work like metric design. PMs view shared dashboards and get automatic coordination emails with other teams if their feature is seen as breaking. People still fight but earlier on and in the same "room" with fewer questions about what's real versus what's noise.
Separating real results from random noise is the meaning of "statsig" / "statistically significant". I think it's similar to how companies define their own metrics (their sense of reality) while the platform manages the underlying statistical and data complexity. The ideal outcome is less DS needed, less crufty tooling to work around, less statistics learning, and crucially, more trust & shared oversight. But it comes at considerable, unsaid cost as well.
Is Statsig worth $1B to OpenAI? Maybe. There's an art & science to product development, and Facebook's experimentation platform was central to their science. But it could be premature. I personally think experimentation as an ideology best fits optimization spaces that previously achieved strong product-market fit ages ago. However, it's been years since I've worked in the "Experimentation" domain. I've glossed over a few key details in my answer and anyone is welcome to correct me.
Hats off to Statsig. They built a stellar product. Superior to many of their industry competitors like Optimizely. Back when I was on an internal Experimentation platform, we were impressed how they balanced dev velocity & stat rigor https://www.statsig.com/updates These guys ship.
Business-wise, I think getting acquired was the right choice. Experimentation is too small & treacherous to build a great business, and the broader Product Analytics space is also overcrowded. Amplitude (YC 2012), to date, only has a 1.4B market cap.
Joining the hottest name next door gives Statsig a lot more room to explore. I look forward to their evolution.
Off-topic, irrelevant question: does anyone need a local first version of Airtable? That uses SQLite under-the-hood and plugs into files and data with syncing across computers.
I’m curious (as a solo dev) if there’s a market for such a product.
Weirdly I’ve been building something along those lines for the last year. Not SQLite backed, but fully local and native (and also does non-local integrations, which you can also script yourself). Should be ready in a month or so if you’re interested!
The relentless demand for justification is indistinguishable from a demand for others to stop doing things you don't like.