Bombos teamBy Piet Baudoin · 30 September 2026
If someone at your company built a custom GPT in ChatGPT, it will stop working after December 11. Today: what you need to do about that, how four people near Rotterdam built a Dutch AI model in a month, and why Anthropic is asking you what you want from AI.
Will my custom GPT still work after December 11?
No. OpenAI is retiring custom GPTs, the assistants you build yourself in ChatGPT. The help page puts it plainly: “Custom GPTs are scheduled to retire on Dec 11, 2026.” That applies to all plans. Only large business customers with an approved deferral get until February.
There is a replacement, a plugin, and you can move your GPT over to it. The instructions and files come along. Conversations do not, and neither does a connection to another program: “GPT custom actions do not transfer through the migration workflow.” OpenAI itself warns that the new version may answer differently.
OpenAI advises you to write down now which GPTs you use and who made them. Save a few questions you often ask them, so you can test the replacement. Why a standalone assistant in ChatGPT isn’t enough for recurring work is covered in can ChatGPT operate your accounting software.
Source: OpenAI Help Center, updated September 2026 and MIXED, 18 September 2026
Can a small Dutch company build its own AI model?
Four IT people at Wainut, an AI services firm near Rotterdam, did it. This summer an American model went offline for a few weeks because of an export restriction. That got them thinking. Luciano Currie told AG Connect, a Dutch tech magazine: “Ons plan was simpel: we trekken er een weekend voor uit, kijken hoe ver we komen.” (Our plan was simple: we’d set aside a weekend and see how far we got.) It became a week, and then a month.
The result is called Walnoot, Dutch for walnut, and it has been freely available since September 28. It is small, Currie says, no replacement for Claude or ChatGPT: “Voor kleine taken werkt het goed.” (For small tasks it works well.) Currie’s example is a familiar one: a letter or email comes in and you want to pull the details out of it. On a European benchmark, Walnoot scores higher than the figures GPT-NL, the Dutch national model, has published. Wainut sells AI services to organizations itself.
What moves me is the tone. Colleague Seleman Arefi: “Laat die talenten nou eens met elkaar samenwerken, is ons idee.” (Our idea is: just let that talent work together.)
Source: AG Connect, 28 September 2026
What is Anthropic asking Claude users?
Anthropic, the maker of Claude, is asking users what they want from AI. An AI asks the questions, in a conversation of about fifteen minutes. It runs from September 29 to October 6. Last year 81,000 people took part. What’s new is that you can make your conversation public, without your name.
Anthropic adds: “How to weigh these benefits and risks shouldn’t be left to AI companies alone.” One of the questions is what you would like AI to change about work, school, healthcare or government.
If you use Claude at work, this is your chance to say what you’re missing. Which tasks already work well is covered in what Claude can do on its own for a small business.
Source: Anthropic, 29 September 2026
What are people saying about it?
A Belgian entrepreneur and investor did the honest math this morning on what AI agents do at one of their companies. The post is in Dutch:
Het werk dat AI Agents dezer dagen in één van m’n bedrijven doet, had minstens twee aanwervingen ‘gekost’.
Maar dat is niet de juiste conclusie. Die stap zouden we waarschijnlijk niet genomen hebben door de onzekerheid van het effect.
Maar als dat effect er wél komt… zal het veel meer aanwervingen betekenen om de opdrachten uit te voeren en de accounts te begeleiden.
Kortetermijndenken is fout hier. AI biedt ontwikkelingskansen, niet alleen aan een fractie van de kost, maar vooral ook zonder de rigiditeit van personeel in België. Als die ontwikkeling tot resultaten leidt, volgen de echte aanwervingen vanzelf. Dat is langetermijndenken, de kern van ondernemen.
@Bert_VdA · 30 September 2026
In English: the work AI agents are doing at one of their companies would have “cost” at least two hires. But that is the wrong conclusion, the post says: they would not have hired for it, because the effect was uncertain. If the effect does come, it will mean many more hires to deliver the work. AI offers room to grow at a fraction of the cost and without the rigidity of Belgian employment rules. Real hiring follows the results. That is long-term thinking, the core of doing business.
And someone at an American hosting company answers Anthropic’s question in advance:
The version of AI I actually want is basically Clippy with manners.
Runs on my device. Private by default. Only touches the cloud when I explicitly say so.
Invisible the rest of the time.
@Mike_demo · 30 September 2026
The same wish twice: AI that works for you, without you losing control of it.
Where are we headed?
An assistant you build in ChatGPT belongs to OpenAI. So it can be gone in two months, as is now happening with GPTs. My expectation: within two years, a small or midsize company will no longer want to lock its AI work into one product from one maker. It will want to know for itself what the assistant does, which rules it follows and where its work lives, so another model can take over.
Bombos takes care of that for you: what your AI agent learns stays with you, not in a GPT that shuts down.
What does this mean for your company?
Today’s common thread is ownership. Of your assistant, of your model, and of your opinion about AI.
Bombos starts with one task. With email into the client file, Bombos finds the right client file for every email with an attachment and puts it in front of you. Our promise: once you approve, everything is where it belongs, and nobody has to save an email by hand anymore. That is the first task, not the whole offer: after that your company does more work with the same people, at a higher quality.
Sources
Every source was opened on 30 September 2026 and every quote appears in it word for word.
OpenAI Help Center: Custom GPT retirement and migration FAQ (updated September 2026) · MIXED: OpenAI is retiring custom GPTs, here’s what carries over to plugins (18 September 2026) · AG Connect: Nederlands AI-model Walnoot scoort hoger dan GPT-NL (28 September 2026) · Anthropic: What do you want from AI? (29 September 2026)