Trusted Intelligence by Aimable

Last time we sat down to record an episode about model independence, our own podcast agent talked us out of it and we made an episode about becoming a self-improving company instead. In the weeks since, Kimi K3 and GLM 5.2 arrived, and the question got more urgent rather than less.Arjé Cahn, our CPO, and Ian Zein, our CEO, on what model independence actually means now that the open weight models are this good.Along the way:* the prospect who looked across the table at Arjé and said: “yes, but those are Chinese models”* why “local” means two completely different things, and only one of them fits on a laptop* why we host in Dublin and Helsinki, and what a hosting company wants from you that a model maker does not* the turning point, and why it rhymes with the week developers stopped writing their own code* building the Excel add-in, and what happened when Kimi read a screenshot of a deliberately broken spreadsheet* three fears about Chinese models, taken one at a time: spyware, coloured answers, and active influence* why a model running on your own hardware is a file rather than an application, and how to prove that to yourself in an afternoon* and the argument that has nothing to do with China at all: if you only have one model, you have nothing to negotiate withIf you take one thing away, make it this: choose the model per workflow, and keep the ability to walk away from it.Trusted Intelligence by Aimable. What’s happening in AI, from the team making it safe to use on sensitive data. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit trustedintelligence.substack.com

We sat down to record an episode about model independence. Our own podcast agent, Esther, had other ideas. She had been reading our recent meeting transcripts, noticed we kept circling a bigger theme, and suggested we talk about that instead. The fact that an agent spotted the pattern, drew a conclusion, and brought it back to us is, when you think about it, the whole point of this episode.Arjé Cahn, our CPO, and Ian Zein, our CEO, get into what it actually means to become a self-improving organization. Not bolting AI onto your workflows, but building loops where AI proposes, people steer, and the work gets sharper every cycle. Along the way:* Jason, our go-to-market agent, and what it is like to work next to a colleague who is an agent* why our meeting transcripts have quietly become more valuable than our documents* the loop we use to turn a conversation into a stronger document, and back again* a maturity model from scattered AI, to governed AI, to your first real loops, and where model independence fits in* why governed AI is the enabler of all of this, not a compliance checkbox* the sameness problem, including the speaker whose AI slides made the audience walk out* and the line that ties it together: good AI is raised, not installed. The same, it turns out, is true of a company.If you take one thing away, make it this: pick your first loop, the one workflow that matters, and keep a human who owns it.Chapters00:00 The topic Esther chose for us02:54 Jason, our go-to-market agent04:55 Transcripts, not documents07:46 When does a conversation become the truth10:47 The loop that makes a document stronger12:46 A maturity model for the self-improving organization14:04 Why governed AI is the enabler, not compliance16:43 Islands, sameness, and putting the humans back in20:59 Good AI is raised, not installed This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit trustedintelligence.substack.com

A Dutch AFM-regulated investment platform processes around 400 deals a year. Each screening memo used to take a full day of manual work - pulling together databases, PowerPoints, financial documents into a templated summary. We built them something that does it in 15 minutes. But the productivity win turned out to be somewhere we didn’t expect.In this episode, co-founders Arjé Cahn (CPO) and Ian Zein (CEO) of Aimable unpack what happened, and what made this engagement different from another “use ChatGPT to summarise stuff” project. The conversation moves from concrete output (a Word document in the customer’s exact template, with an Excel sheet underneath) to a broader question that keeps showing up across Aimable’s customers: how do you make AI deliver *finished work* instead of just talking about work? And what kind of company do you have to be to do that?Along the way they cover why an LLM can’t be trusted to calculate but a skill wrapping Python around it can, why a black marker doesn’t work as redaction, what they learned from Ian’s own accountant chasing a missing Coolblue invoice, and where this is heading once agents start talking directly to other agents.0:00 The 15-minute investment memo0:30 Inside the customer engagement2:30 Finished work vs ChatGPT experiments3:50 One-off prompts vs reusable skills5:00 When the numbers have to be right7:00 An accountant’s email - and a much smaller skill9:30 Why funds don’t just drop pitch decks into AI10:00 A different kind of software company11:30 The black-marker problem (smart redaction)13:30 The compounding platform14:00 Portal visions and agents that play by the same rulesAbout Trusted IntelligenceTrusted Intelligence is a weekly conversation between Aimable’s founders about what they actually run into while building a safe AI platform for organisations that need their AI to deliver real, grounded work - not chat.About AimableAimable is the leading European AI platform that lets organisations get real work done with AI, on their own data, in their own format, with their own rules - and the audit trail to prove it. Standard platform, customisable skills, EU-hosted by default.🔗 www.aimable.ai This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit trustedintelligence.substack.com

We gave our AI agents real jobs. At home and at work. This is what happened.Arjé’s kids now negotiate their chores with an AI called Rosie over WhatsApp. One insisted on being addressed as “The Untouchable Alpha Lord of Chill.” Ian’s agent Zosia sends him a daily briefing at 9AM. His contractors thought she was a real person.Then we brought it to work. Aimee, our company AI agent, reads our Slack channels, analyzes meeting transcripts, and delivers a briefing at 5AM that connects dots we missed ourselves. She does the work of at least three people. And 99% of companies aren’t doing anything like this yet.But it’s not all smooth. These agents can’t stop talking, even when you tell them to be silent. Aimee still reports every morning: “I will NOT say anything about Arjé’s to-do list.” That’s funny at home. At work, with customer data involved, it’s a real problem. And that’s where safety gates come in.In this episode we share what actually works, what fails, and what mid-market companies should be thinking about before their employees start running agents on their own.Trusted Intelligence is a podcast by Aimable, the AI productivity platform for organisations with sensitive data. We build in the open. This is what that looks like.00:00 Meet Zosia and Rosie: our personal AI agents03:15 Rosie manages the family chores via WhatsApp07:15 One Mac Mini, two agents, strictly separated07:45 Aimee: our company agent that never sleeps10:25 Human + AI collaboration: more than an RSS feed14:05 “99% of companies aren’t doing this”15:45 When contractors think your AI is a real person16:20 Why AI agents can’t shut up (and why that matters)18:50 From personal to family to work: the progression20:05 The cost of agents that run 24/7 This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit trustedintelligence.substack.com

Stop Apologizing for AI: Build Strong Core Documents and Iterate with TranscriptsTwo founders discuss why people still apologize for using AI—often to avoid seeming lazy or to preempt mistakes in generic AI output—and note visible tells like default slide layouts and em dashes, which they no longer try to hide. They explain how to set an “AI for everything” expectation inside a company, including onboarding a non-technical sales hire by giving him core documents and an AI that can generate needed assets instead of preparing slide decks manually. They argue AI produces “garbage” when given no business context and that prompting is a skill, but a simple approach is to create concise, aligned value proposition and brand voice documents through an iterative process: record round-table discussions, transcribe locally (e.g., iPhone Voice Memos plus Whisper), and feed transcripts back to AI to refine. They emphasize owning AI outputs and integrating iteration inside the AI workflow.00:00 Why Apologize for AI00:51 Owning AI Artifacts01:39 Setting Team Expectations03:28 Why AI Outputs Feel Generic04:14 Build Solid Core Docs05:34 Roundtable Recording Method07:46 Simple Tools and Privacy09:52 Brand Voice Playbook11:29 Iterate and Constrain Models14:14 Concise Docs Win16:06 Own the Output17:25 Wrap Up and Next Time This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit trustedintelligence.substack.com