Sovereign Stack

On this episode of the Sovereign Stack Podcast, Stephane talks with Abbas, Chief Legal Officer at Huck, about practicing law now that AI is part of the job. They start with some of the strangest cases and laws from around the world. Then they move on to what AI means for lawyers, clients and the legal profession.

What they cover

• Defending clients: Why everyone has a right to a defense, and why a lawyer can still refuse a case they can't take on morally.
• Law in Lebanon vs. abroad: How paper-based courts slow justice, and why civil law and common law systems can't be compared like for like.
• AI makes good lawyers better: Abbas argues AI will make a good lawyer exponentially better and a bad lawyer worse. Without foundational knowledge, you can't judge the output.
• Wrapper vs. proprietary engine: How a thin interface over a general AI model differs from a purpose-built legal engine. He explains what he calls the "AI paradox": more data can mean less accuracy, which is why ontology and sovereignty matter.
• Billable hours: Faster work is pushing firms toward lump-sum fees or higher hourly rates for fewer hours.
• Regulation and ethics: Why lawyers shouldn't wait for rules to use AI ethically, and why regulatory sandboxes help.
• Limits of AI: Where your data goes matters, and a skilled lawyer must stay in the loop.
  1. Where the profession is heading:
  2. New specializations such as AI governance and data protection.
  3. Firms that use AI grow, small firms merge, and firms that don't use it fall behind.
  4. Over-reliance on AI by young lawyers is a bigger risk than AI itself.

His advice to law students: Build your legal knowledge as if you had no access to AI, then use AI to multiply it.

n this solo episode of the Sovereign Stack Podcast, Stephane records a voice memo from a balcony at his office. He thinks out loud about where AI is heading and why everyone should have a personal AI of their own.

Starting from a simple question, "how do I get real work done from my phone?", he explores:

• The limits of mobile AI: Today's phones can send commands to AI tools, but they can't yet replace a laptop. Hardware, screens and interfaces still need to catch up.
• Local models and privacy: The case for running a model on your own device, trained on your own context, with your data staying with you. He also looks at what that could mean for Siri, Alexa and ad-driven business models.
• Personal sovereign AI: The idea that sovereign AI isn't only for companies. Individuals should have a native AI that builds securely for them alone.
• "Your tool," not "internal tool": Build for yourself first. If it works, consider offering it to others.
  1. How to start learning AI now:
  2. Build a "second brain" that captures your conversations and context.
  3. Automate daily routines like email sweeps and morning briefs.
  4. Go further with quarterly reports and financial dashboards.
• Filtering the noise: Learning by doing helps you tell copycat tools from genuinely useful ones.
• Human in the loop: AI makes good choices fast, but you make the final decision.

He also explains why he doesn't edit this podcast: it's meant to be a human voice, not a polished studio production. His advice is to learn from each other and use AI to give yourself a more efficient life and a better work-life balance.

Stephane talks with Nicolas, community manager at HAQQ, who taught himself AI video production without a video background. They explain the process behind the AI videos HAQQ now publishes. Their goal is to avoid the generic, one-shot look people call "AI slop."

What they cover:

• Why code-to-video is taking off: Large language models are good at writing code, so they can turn plain-English ideas into animated, watchable video.
• Hyperframes vs. Higgsfield: Hyperframes works as the editor that sequences and animates. Higgsfield generates the assets, routing to different models for images, audio and video. They work together rather than compete.
• The Picker: A review page where the team chooses between AI-generated versions of each frame. It lets them check consistency before spending time on a full render.
  1. Their playbook:
  2. Start with a story.
  3. Plan a frame-by-frame version.
  4. Generate cheap static images first.
  5. Turn those into GIFs.
  6. Build the full video with transitions.
  7. Add music and edit.
• Better prompts: They write detailed prompts in a task manager, covering story, characters, colors and style, instead of sending a raw idea.
• Testing before posting: They try content as Instagram trial reels, then post the winners on other platforms.
• Humans and tools: Taste, storytelling and direction still need a person. Motion graphics tools like After Effects are increasingly replaceable with code.

Their advice: build your taste by consuming good content, keep a person in the loop at every step, and focus on quality, speed and consistency.

In his first solo episode of the Sovereign Stack Podcast, host Stephane Boghossian shares his screen and walks through how he works with AI day to day. The setup centers on Claude Projects, Claude Code and artifacts, plus Aleph, the system he built to hold his knowledge base and automations.

What I cover:

• Projects and workflow: How he splits his work into projects and threads. He sends ideas, notes and Instagram links to the right project at night, then reviews the finished artifacts and tasks in the morning.
• Artifacts as a learning tool: Interactive graphs and visual explainers he built to understand topics like the Anthropic IPO and frontier models. He also made a full hardware and GPU course with a 100-question exam.
• His "second brain": A knowledge graph built from his meeting transcripts and AI sessions. It lives on his own machine and in a private repository.
• Work artifacts: A centralized sales materials hub with PDFs in Arabic and English, a legal taxonomy, a database of Lebanon's history, and redesigns of the product onboarding and chat.
• Daily rituals: An 8 AM briefing that tracks news, stocks and competitors, plus an AI trends explorer. It maps 51 clusters, 301 micro trends and 27 sources.
• Human in the loop: He reviews and approves what the system produces and checks its claims before anything ships.

His message is that you don't need to copy his setup. Build your own system, because the people going this deep into automation and AI agents are about 2% of the population, and this show is for everyone else.

In this episode of the Sovereign Stack Podcast, Stephane talks with Ronald Harfouche, VP of Sales at HAQQ. Ronald was previously a chief revenue officer across several regulated industries. They look at how a sales team can use AI when client data is sensitive.Ronald explains how the team brings in leads and turns them into meetings and customers:

• Demo bookings go straight to a tailored demo, built around each prospect's use case.
• Free sign-ups are scored on factors like credit usage, email type and stated profession. The best ones are exported to Apollo and placed into outreach sequences by email, phone and LinkedIn.
• Outbound lists are built by company, location and persona, then enriched with Apollo contact data.
• Salesforce syncs with Apollo and uses enterprise AI agents to log meeting notes and update next steps.

They also cover:

• How AI helps sales teams quickly customise collateral, proposals and presentations with a client's logo and use case
• Why the team keeps strict limits on what AI can connect to, to protect client emails and phone numbers
• Where AI helps salespeople (prioritising leads, refining messaging) and where humans still matter, especially on calls and voice

Stephane sits down with Jean, who recently moved from operations to growth at HAQQ. They talk about how he went from using AI as a "Google++" to building his own automations.Jean has worked on more than 300 events over six years. He explains how he built an AI scoring system that rates each event out of 100 on location, timing, speakers, agenda and other factors. Out of about 100 events for 2026 and early 2027, it helped narrow the list to 4–5 events to speak at or exhibit at, and about 10 to attend.In this episode:

• Why Jean joined HAQQ and how his role shifted from ops to growth
• How the event scoring system works, plus a story of a two-hour trip to meet one key person
• Why AI can recommend, but humans must make the decisions and show up
• How to choose which repeating tasks to automate, and when to stop iterating (set clear KPIs first)
• Why he approves each automation step instead of letting the tool run on its own
• Stephane's 5-step framework for starting with AI:
  1. Identify the task to automate.
  2. Map the possible steps.
  3. Define the output.
  4. Gather context and research.
  5. Prepare your prompt and monitor the conversation.
• Jean's two biggest lessons: AI helps you do things you couldn't do manually, and the time it saves lets you learn things outside your expertise