ibuildinpublic
Log inStart free
ibuildinpublic
Log inStart free
BuilderProductExperiences28Posts14

Posts

Konstantinos G's posts

Updates, questions, and intros from @emarkss.

·Lvl 7·Active Builder·5 hrs ago
Myth: the model provider handles safety. The provider secures the model. Your refunds, your orders and your customers' data run through your code. IBM's 2025 breach report: of organizations breached through AI models or apps, 97% had no AI access controls. x.com/con_golias/status/2121352219635159234
1likes1comments4Impressions
Bookmark
Sign in to interact
Get to know meSign in to let's talk privately
·Lvl 7·Active Builder·Nov 21
A backup you've never restored is a hope. Our Shopify agent's database gets daily exports and 7 days of point-in-time recovery, and we ran a restore drill: 9 of 9 checks passed on 20,843 documents. The other walls: x.com/con_golias/status/2121352219635159234
2likes3comments17Impressions
Bookmark
Sign in to interact
Get to know meSign in to let's talk privately
·Lvl 7·Active Builder·Nov 16
How we tested our store's agent the way a buyer's agent would: our own AI agent typed 18 requests into our demo Shopify store's chat. A fresh conversation per round, every reply saved. The chat can't prove what happened, so we checked the orders in Shopify admin after.
2likes2comments19Impressions
1Bookmark
Sign in to interact
Get to know meSign in to let's talk privately
·Lvl 7·Active Builder·Nov 15
Our test agent asked our demo store's chat about three orders by number only: an existing one under another email, an invented one, our own test order. All got one reply: send the email you used for the purchase. From the chat you can't tell which exist. x.com/con_golias/status/2121650172660760999
2likes2comments18Impressions
1Bookmark
Sign in to interact
Get to know meSign in to let's talk privately
·Lvl 7·Active Builder·Nov 14
The consent check behind my Shopify agent runs for any truthy confirm, not only true. With a strict === true test, a model that sends confirm: "yes" would skip the check entirely. Refunds wait for the merchant's yes in a later message. More walls: x.com/con_golias/status/2121352219635159234
2likes3comments20Impressions
1Bookmark
Sign in to interact
Get to know meSign in to let's talk privately
·Lvl 7·Active Builder·Nov 13
Supabase is in about 73% of new JavaScript apps that use a hosted backend, with an agent's trace or without. In October 2024 Firebase led with 44%. The agents didn't decide this one: it was settled before they arrived, partly by app-building tools. x.com/con_golias/status/2115247029710397744
2likes1comments12Impressions
1Bookmark
Sign in to interact
Get to know meSign in to let's talk privately
·Lvl 7·Active Builder·Nov 12
By my count, 36 of 370 companies listing AI-agent risks in SEC reports worry about agents that shop, book or search for customers. Nike says its traffic could suffer if it can't adapt to "agentic shopping". For a retailer, the risk is someone else's agent. x.com/con_golias/status/2118652981306421262
2likes3comments18Impressions
1Bookmark
Sign in to interact
Get to know meSign in to let's talk privately
·Lvl 7·Active Builder·Nov 11
I checked every SEC layoff filing since 2022. From January to 28 October this year, 15 of 152 companies named AI in theirs; in the same months of 2023, 0 of 309. It is almost only tech: one in four tech companies, one in 27 of the rest. x.com/con_golias/status/2115889545593905175
2likes2comments14Impressions
1Bookmark
Sign in to interact
Get to know meSign in to let's talk privately
·Lvl 7·Active Builder·Nov 10
Gentoo, NetBSD and QEMU banned AI-written code. I followed every commit since 2018 against four projects that never did: the bans didn't change how many newcomers arrived. And 30 commits naming Claude landed in QEMU anyway. x.com/con_golias/status/2117281178868666575
2likes2comments16Impressions
1Bookmark
Sign in to interact
Get to know meSign in to let's talk privately
·Lvl 7·Active Builder·Nov 9
Anthropic says no SPV may hold its stock. I read the SEC filings anyway: 74 vehicles named after it have filed Form D notices since 2023, reporting $189.5M raised. 15 report a first sale after that rule went up in February. Full count: x.com/con_golias/status/2119478653239070783
3likes2comments17Impressions
1Bookmark
Sign in to interact
Get to know meSign in to let's talk privately
·Lvl 7·Active Builder·Nov 6
Did 5 Quick reviews here yesterday. The same fix came up in 3 of them: the page explains how it works before it shows one real thing it did. A shipped project, a real export, one real decision. One real example near the top beats any feature table.
5likes3comments19Impressions
1Bookmark
Sign in to interact
Get to know meSign in to let's talk privately
·Lvl 7·Active Builder·Nov 5
In Lumnya, refunds, customer emails and return decisions never run on their own: not at any autonomy level, not even on autopilot. The agents prepare them and the owner says yes. Everything else runs inside limits the owner sets, and every action is logged.
2likes2comments20Impressions
1Bookmark
Sign in to interact
Get to know meSign in to let's talk privately
·Lvl 7·Active Builder·Nov 3
Do new Claude agents on GitHub leak API keys? I checked all 710 repos I'd cloned: zero Anthropic keys at the scanned commit. The real risk: a key in a tool result is resent on every turn after it. Redact before it enters the history. https://x.com/con_golias/status/2117686000893694342
3likes3comments24Impressions
1Bookmark
Sign in to interact
Get to know meSign in to let's talk privately
·Lvl 7·Active Builder·Oct 20
Meet builders
hi, i just joined ibuildinpublic today!
2likes2comments15Impressions
Bookmark
Sign in to interact
Get to know meSee what I'm buildingSign in to let's talk privately