Gen Vagula

Jensen Huang with a Chinese spy and Jensen Huang with a Russian spy.

With Optimus @elonmusk will solve the issue of sending someone to the long journey to Mars. Team of Optimus robots would be those first astrounaouts.

Fan of ATC and aviation? Want to track airceafts? Check out our Ampron Sky webapp. Lot of professional features and analytic options. sky.ampron.eu

Everyone keeps talking about how AI is going to take our jobs. For months now, I’ve been trying to offload my work onto an entire army of AI agents, but damn it, they just won’t take it! I’ve tried everything, the carrot and the stick, but they simply refuse to do my job. I still end up having to do it myself. On top of my own work, I now also have to babysit them and keep prodding them along. And to make matters worse, I end up having to do extra tasks that they assign to me. So, spare me that talk. Or maybe I’m just doing something wrong.

Did a writeup about AI memory topic. AI agents start every task with no memory of the last one. I explain why, using the film Memento, what 2026 research says, and the note architecture we are moving to at Ampron. Read article: genvagula.ee/en/writing/meme…

An assistant is something you ask things of. A worker is something you give tasks to. A director is something you give a domain to. genvagula.ee/en/writing/tool…

Lately I have been following a debate among the heavyweights of Estonia’s technology world about creating an ID code for AI agents. The idea is that artificial intelligence could act safely on a person’s behalf, with clear accountability. A necessary thing, probably. But if that is our big ambition for the AI era, we are flying very low. In my view the Estonian state should set itself a much bigger goal: to build its own national large language model. Read more: genvagula.ee/en/writing/esto…

Read what we built with our Claude Developer ampron.eu/blog/ampron-sky-li…

An AI executive should not have to carry thousands of historical conversations, operational records, and previous decisions in its active context. It should have reliable access to them.

Olen jälginud viimasel ajal Eesti tehnoloogiamaailma kõige vägevamate poleemikat AI-agentidele agendikoodi loomise üle. Et tehisaru saaks inimese nimel turvaliselt tegutseda ja vastutus oleks selge. Vajalik asi ilmselt. Aga kui see on meie suur ambitsioon AI-ajastul, siis lendame ikka väga madalalt. Minu arvates peaks Eesti riik seadma endale hoopis suurema eesmärgi: ehitada oma riiklik suur keelemudel. Mitte kohandada USA tehnoloogiat, vaid luua riigiettevõttes võimekus oma keelemudelit arendada ja hallata. Selleks peaksime minu hinnangul võtma sihiks vähemalt miljard eurot aastas. Mitte ühe mudeli treenimiseks, vaid kogu võimekuse jaoks: arvutuskeskused, turvaline taristu, andmed, tippspetsialistid, teadus ja rakendamine riigiasutustes. Jah, miljard on meie mõistes tohutu summa. Kuid kui AI hakkab tegema järjest suuremat osa meie tööst, ostame vastasel juhul selle kriitilise võimekuse väljast sisse. Rahastus peaks käima etappide kaupa, mõõdetavate tulemuste ja selge vastutusega. Osa investeeringust võiks pikemas vaates tagasi tulla riigi säästude ja teenusetuluna. See oleks julgeoleku- ja majandusprogramm, mitte üks tarkvaraprojekt. Ükskõik kui kehvake meie esimene keelemudel oleks, oleks see meie enda oma. Meie kontrolli all. Meie arendada. Meie vajadustest lähtuv. Esmalt saaksid seda kasutada riigiasutused tööprotsesside automatiseerimiseks, siis Eesti ettevõtted. Kui lahendus osutub konkurentsivõimeliseks, saaks seda pakkuda ka teiste riikide ettevõtetele ja valitsustele. Siin on üks majanduslik aspekt, millest räägitakse liiga vähe. Üha suurem osa tööst, mille eest seni maksti inimestele palka, tehakse ära AI abil. Aga kellele maksame selle AI eest? Suures osas USA tehnoloogiaettevõtetele. Osa rahast, millest varem maksti Eestis palku ja laekusid maksud, võib tulevikus liikuda välismaale AI-teenuste eest. See puudutab meie tulevast maksubaasi, majanduslikku sõltumatust ja julgeolekut. Oma keelemudelit saaksime pakkuda tasulise teenusena Eesti ja välismaa ettevõtetele. Ainult ostja asemel oleksime ka tehnoloogia omanikud ja teenusepakkujad. Ja mis kõige olulisem: me ei oleks nabanööri pidi seotud mõne üksiku USA tehnoloogiaettevõtte otsuste, hinnastamise ja teenuste kättesaadavusega. Mida sügavamale AI majandusse ja riigivalitsemisse jõuab, seda ohtlikumaks selline sõltuvus muutub. Ma ei arva, et Eesti suudaks homme OpenAI, Anthropicu, Meta või Google'iga võistelda. Aga oleks strateegiline viga jätta oma võimekus arendamata lihtsalt seetõttu, et teised on praegu ees. Meil on digiriigi ehitamise kogemus ja tehnoloogiline kompetents. Nüüd oleks vaja sellele vastavat ambitsiooni. AI-agentidele isikukoodi loomine on digitaristu arendamine. Oma keelemudeli loomine oleks investeering Eesti tehnoloogilisse iseseisvusse. Me ei peaks püüdlema ainult selle poole, et olla maailma kõige osavam võõra AI kasutaja. Peaksime tahtma olla ka selle tehnoloogia omanik ja arendaja.

Unpopular opinion, Google destroyed the internet with their Google ads program. Most of the pages got toxic with some of them bad clickbait ads that were luring people to click and others were just massively cluttering the pages. Site owners choose money over user experience. Now they got no money and no users.

What reading material you would advise i should give to my AI executive agents? For their education as a leaders.

Anyone else frels that Claude Code llm is underperformkng today? Opus 5.5 has been like low iq agent making mistakes.

For all hobby developers. We do not have to visit this page ever again. If @AnthropicAI or @OpenAI have done anything good in the world, this is one of those things! The stress we have got from that site has been irrecoverable.

How our office runs at Ampron with AI Executives. The future is now. They are enthusiastic bunch of new coworkers that push us further and help us grow the company. #ampron #ai #office ampron.eu/blog/we-automated-…

Significant changes. Now regular people can see planes with aircrafy icons. We try to balance this to be suitable for experts and just the regular people who want to check the flights. Switch is in the settings to get the expert mode to rectangle icon. #ampron #aviation

Every post i see bashing vibe coders, are made by software developers. Coping is really hard.

I wish the bots would go away from X like was just a week ago. These were so beautiful times over here in X.

It is sad that @elonmusk does not use his own AI agent. Could be much smarter in some areas. I asked about his leaning towards critisizing Ukraine and not much towards Russia. Grok brought in some valid neutral points, have to agree with him. But could not find a single post where he critisizes Russia.

This guy gets it. Imagine being smarter in strategy than the leader of the biggest country in the world.