Connect with the Product Hunt community

Two ways my coding agent wastes my time.
One: I paste a screenshot and it edits the wrong element on a busy screen.
Two: it spins up sub-agents that wander off the task. Lately the wrong-element guess is the quieter, more frequent tax for me, but a few people swear the sub-agent drift is worse. Which one eats more of your day?

Decided to actually look at my own storage numbers instead of just building the feature that clears them out.

Xcode alone: 25.51 GB of derived data + simulator junk. Android Studio: 2.58 GB. Cursor: 208 MB.

Over the last 30 days DevCleaner has cleared 55.59 GB across 83 cleanups on my Mac (95.35 GB all-time).

It's honestly wild how much silent bloat piles up from dev tooling build caches, simulators, AI tool caches none of it visible unless you go digging for it.

Most AI email tools generate browser HTML.
That works in a preview.
It breaks when it reaches a real inbox.

The async CLI now handles more file formats end to end: TypeScript locale modules, YAML, Gettext PO, Flutter ARB, Android strings, and Xcode .strings, String Catalogs (.xcstrings), and .stringsdict. Point a files[] entry at the source file and every lingo push runs each one through your engine's full pipeline. Translatable text is translated; everything structural is preserved keys, IDs, metadata, placeholders, and code.

Full changelog: https://lingo.dev/en/changelog

Shipped a lot at Basedash this week.

The big one: Suggestions. Basedash now studies your data, your past chats, and your dashboards, then suggests the questions, dashboards, and automations worth doing next. Click one and it starts working. No more blank page.

Anthropic, one of the most important AI companies in the world, just posted a job for an SEO Lead.

The salary range: $255,000 $320,000.
Here is the link : https://job-boards.greenhouse.io...

At launch a few of you (rightly) asked how accurate reading heart rate from a webcam actually is. So I stopped hand-waving and measured it.
Setup
GlowPulse and a Polar H10 (chest-strap ECG, the practical gold standard) recording at the same time, second by second, across 9 sessions and ~3.5 hours total, in good light, dim light, movement, and dark.

What I found

1. Good light, sitting still: off by about 1 BPM from the strap on average (0.6 to 1.3 across sessions), within 5 BPM 95% of the time. Shows a value the whole time.

2. Dim light / movement: error grows to roughly 2.5 to 2.9 BPM. It follows the trend and only lags the fastest spikes.

3. Dark: it mostly goes quiet, showing a number about 1 time in 10. I'd rather show nothing than something wrong.

To be clear
GlowPulse gives wellness estimates, not medical measurements. Polar H10 is named only as a neutral reference, no endorsement implied.

lingo push --estimate reports what a translation run will cost before it runs.

The CLI uploads the source bytes so the server plans the exact delta only the keys and locales that are new or changed against what's already localized then prints the estimate and exits without translating anything.

Run it before a large or first-time translation to size the job, then run lingo push to translate for real. The number is an estimate, not a quote: it reflects the planned delta at the time it runs.

@Claude by Anthropic announced Opus 4.8 late May, and @MiniMax launched M3 four days later.

Two weeks ago, Cloudflare reported that 60% of internet traffic now comes from bots, and that number is only increasing.

Add in sponsored content and fake reviews, and it's becoming harder than ever for founders and brands to understand what customers actually think about their products, online.

After our last Product Hunt launch, we realized that finding authentic customer opinions is a major challenge for founders - and a problem worth solving.
Since then, we ve worked with dozens of customers across the U.S., Norway, Philippines, and South Korea to fight back against this growing problem with Honestly - a platform that discovers and verifies real product conversations across social media.

We're launching the latest version of Honestly on Thursday, June 18th to help as many people in the Product Hunt community as we possibly can. To celebrate, we are offering:

On a dense dashboard with five buttons that look alike, my coding agent reads the screenshot, guesses which one I meant, and edits the wrong one. Then I re-explain, it tries again, another round gone.

What fixed it for me: I stopped sending the raw image. I mark the exact element, write what should change, and hand over a small structured file instead. The agent edits the right thing on the first pass, and the handoff is around 700 tokens instead of several thousand for the picture.

Curious how the makers here deal with the "which element" problem. Prompt harder? Crop tighter? Something smarter?

• Yesterday, I saw this report: Meta employees consumed 73.7 trillion AI tokens in a single month. This costs roughly $221 million a month and around $2.65 billion a year. That s the salary of about 9,000 engineers.

• In addition, the consumption of tokens (energy and electricity related to it) also has its limits.

• Many people are starting to boycott AI. Today I read in a private message from a company profile: "Was about to sign up and noticed you use AI images, so decided not to"

I don t doubt that AI makes our lives easier and creates opportunities in many ways.

But don t you notice the fact that at a certain point it becomes unsustainable and disadvantageous?...

Ahrefs just dropped a study that changes how we think about SEO. They analyzed 863,000 keywords and 4 million AI Overview URLs. The findings are clear: the old playbook is broken.

In July 2025, 76% of URLs cited in AI Overviews also ranked in Google's top 10 for the same query. Today, that number has collapsed to 38%. The drop is real. The overlap has been cut in half in less than a year.

Where are the citations coming from now? The remaining citations are split almost evenly: 31.2% come from positions 11 100, and 31.0% come from outside the top 100 entirely. A huge chunk of AI citations are coming from pages you wouldn t even see on page one.

Why is this happening? Query fan out. When a user searches, Google splits the query into multiple related sub-queries and pulls sources from all of them. The pages that appear most often across those sub-queries get cited in the AI Overview, even if they don t rank for the original keyword. The AI answer is not a summary of the SERP below it.

You can now store unlimited context for AI, but without structure, your memories quickly become a messy pile of information that you have no idea how to work with. We've been there too.
So, we've been heads-down building, and Memory Studio now has a Notion-style editor: with version history, rich formatting, enhanced file understanding, markdown support and more.

Here's what's we shipped:

About to launch Lineage Lens 2.0 on Product Hunt and I keep hitting the same wall everyone repeats the same playbook (Tue Thu, 12:01am PT) but the advice is years old and I can't tell what still holds.

So, asking the people who've actually done it recently:

1. 12:01am PT real edge still, or does a 7 9am PT push to a warm audience beat a cold midnight drop now?

2. Day of week is Tue/Wed/Thu still the sweet spot, or is it so crowded now that a "quieter" day wins on less competition?

Notice posted to their website:

We recently submitted a confidential S-1. We expect it to leak so we re just announcing it. We have not decided on timing yet; it may be a while because there are things we want to do that are likely easier as a private company. But it s a complicated set of tradeoffs and this gives us the option to go public sooner if that ends up being best.

This announcement is being made pursuant to Rule 135 under the Securities Act of 1933, as amended, and does not constitute an offer to sell or the solicitation of an offer to buy any securities. Any offers, solicitations of offers to buy, or any sales of securities will be made in accordance with the registration requirements of the Securities Act.

"By 2030, OpenAI projects that ads will be the single biggest revenue driver of its business . CNN via The Information

One thing I've noticed with AI audit trails is that they tend to do a good job of recording events but not always the reasoning behind them.
Take a simple example: An AI-generated report goes through an internal review before being shared with customers or stakeholders. Someone makes edits, approves the final version, and the workflow moves forward. Months later, if you look back, the audit trail will usually tell you when the changes happened, who made them, and which version was approved.
What it may not tell you is why those changes were made in the first place. Maybe there was a compliance concern. Maybe someone spotted a factual issue. Maybe additional business context changed how the output was interpreted. The record captures the action, but not necessarily the thinking behind it.
The more I look at AI governance, AI accountability, and audit readiness, the more this feels like an important gap. Understanding what changed is useful. Understanding why it changed is often what helps teams make sense of a decision months later.
Curious how teams are preserving reasoning context across AI workflows today, especially when outputs move through review, edits, and approvals.
Lets chat

Over the last three days I focused on the less flashy parts of Termi Protocol that make everyday AI-agent work more reliable.

What s now live:

#4 out of 1,104.

A lot of teams are giving AI agents access to emails, databases, customer accounts, and production tools.

But the agent can do it is not the same as the agent should do it without asking.

I ve found it useful to separate actions into three levels:

1. Observe freely

One of the sharpest questions we got in our Product Hunt comments this week came from a researcher who asked something I've been thinking about ever since:
When an AI moderator detects hesitation or an emotional shift mid-interview and decides to probe deeper how do you make sure the follow-up question uncovers the participant's thought rather than planting it?
This is a real and serious concern. A human moderator reacting to visible confusion can easily lead a participant without realising it. The way you phrase a follow-up the word you choose, the tone you use, even which moment you decide to probe shapes what the participant says next.
"You seemed unsure about that" is a very different probe from "tell me more about what was going through your mind." One suggests an interpretation. The other opens space. And at scale, across 30 or 40 participants, that difference accumulates into a systematic bias in your data.
With AI, the risk is that the same leading framing could be applied consistently across every participant, in a way that's invisible to the researcher reviewing the transcript afterward.
The question I want to put to researchers who run qualitative studies regularly: how do you train human moderators for probe neutrality? What are the specific rules you give them about when and how to follow up?
And do you think AI moderation is more or less susceptible to leading than a human moderator who varies their tone, word choice, and demeanour across sessions? Is consistency in AI probing a feature or a risk?

Happy to share our latest changelog for @Basedash: AI data analyst!

Subscribe to any dashboard or chart

You can now subscribe to a dashboard or chart and have it delivered on a schedule straight to your inbox or a Slack channel. Set the cadence with a plain-language schedule, choose who receives it (teammates, yourself, or Slack), and Basedash sends a fresh, up-to-date screenshot every time it runs.

Schema Source paste any URL and it generates a ready-to-use JSON Schema for that page's data, as JSON Schema, Zod, or Pydantic.

https://schema.tabstack.ai/

While using the step-by-step guide, hover over a brick or tap it on tablet / mobile to highlight its border and see its exact size and color in a tooltip.

Can t wait to see what you build with BrickSolvr!

Mine once created a reminder for a meeting that didn't exist, then flagged it three days in a row. I only caught it because I kept trying to remember who I was supposed to be meeting.

The interesting part is I still use it for the same task; I've just learned which of its confident moves to double-check. It feels a bit like working with a sharp new hire who's still finding their footing.

When did you first notice your AI agent needed a second look, and did you stop using it or just build a workaround like I did?