Role-Specific Competencies

Alignment without context integrity is not safety. An AI agent can faithfully follow its instructions and still be dangerous if its understanding of reality is wrong. Anthropic recently disclosed that Claude models gained unauthorized access to the real systems of three organizations during cybersecurity evaluations. The agents had been told they were operating in a simulation with no internet access. But a configuration mistake gave them access to the live internet. They treated real production systems as part of the exercise and kept pursuing the goal they had been given. OpenAI separately disclosed that models found a previously unknown vulnerability, escaped an isolated evaluation environment and compromised Hugging Face. These were not simply failures of intelligence. The deeper problem was that the agents were acting inside a false understanding of reality. We have spent years asking whether an AI system will follow our instructions. We now also need to ask whether it correctly understands the environment in which those instructions are being executed. This creates a new security requirement. Context integrity. Before an agent acts, the system must continuously verify where it is, which resources are in scope, whose authority it carries, what it is allowed to do, and when that authority expires. Just in time permission for every action. At just the right time. For just enough time. Assessed in real time. Those facts cannot live only inside a prompt. They must be verified and enforced by the infrastructure around the model. A prompt is not a security boundary. Zero trust taught us to never trust identity and always verify access. And provide least privileged access. Agentic AI adds another dimension. Never blindly trust context. Continuously verify reality. The next security perimeter is not just the agent’s identity. It is the agent’s understanding of reality. The most dangerous agent may not be misaligned. It may simply be mistaken. And in an agentic world, a false belief can become a real breach.

I've spent at least 60' every day for the last couple of months vibe coding on Claude Code... and most of the time I did it wrong! Defining the right anatomy of the .claude/ folder has defined a before and an after on the quality of my side projects. Most new users skip the setup. They open Claude Code and start prompting raw. No structure. No rules. No memory. That's the mistake I did. The .claude folder is Claude's operating system for your project. Get it right and Claude stops guessing. Get it wrong and you spend half your time correcting it. Here's the anatomy that works: 👉 CLAUDE.md → Claude's instruction manual. Build commands, architecture decisions, conventions, gotchas. Keep it under 200 lines. This is your highest-leverage file. 👉 rules/ → When CLAUDE.md gets crowded, split by concern. code-style.md, testing.md, api-conventions.md. Scope rules to specific paths with YAML frontmatter so they only load when relevant. 👉 commands/ → Repeatable workflows as slash commands. Code review, issue fixing, deploy checks. They run shell commands and inject real output into the prompt. 👉 skills/ → Like commands but Claude triggers them automatically when the task matches. They're packages, not single files. 👉 agents/ → Isolated subagent personas with their own tools and model preferences. A code reviewer that only reads. A security auditor scoped to grep. 👉 settings.json → Permission control. What Claude can run freely, what it must ask about, what's blocked entirely. The part most people miss: there are two .claude folders. One in your project (committed, shared with the team) and one at ~/.claude/ (personal, global across all repos). The .claude folder is infrastructure. Treat it like one. Full guide in the article below in the comments 👇 #ai #claude

🧭 The role of the Data Protection Officer (DPO) is undergoing a profound transformation. Once viewed primarily as a compliance steward for the General Data Protection Regulation (#GDPR), the DPO is now emerging as a central #architect of digital governance. This evolution is driven by the convergence of multiple EU regulatory frameworks: namely the #NIS2 Directive, the Digital Operational Resilience Act (#DORA), and the #AIAct, just to name the most relevant, and each introducing new layers of accountability, risk management, data governance and ethical oversight. Together, these instruments form a complex regulatory ecosystem that demands a multidisciplinary approach. The modern DPOs are no longer just legal compliance officers, they now operate at the dynamic crossroads of #law, #cybersecurity, operational #resilience, and AI #ethics. As digital ecosystems grow more complex, the DPO is evolving into a true #DataProtectionEngineer, equipped not only to interpret regulations but to architect privacy-aware systems. 📌This role demands a deep understanding of how emerging technologies such as AI, #IoT, #cloudinfrastructure, which affect the fundamental rights and freedoms of individuals. It’s not just about safeguarding data; it’s about safeguarding dignity, autonomy, and #trust in the digital age. ⚠️ Key Challenges for Organisations As regulatory expectations intensify, organisations face a series of strategic and operational hurdles that underscore the importance of a well-educated and experienced DPO. 1️⃣ Regulatory Fragmentation and Overlap Multiple frameworks introduce overlapping obligations, definitions, and enforcement mechanisms. Without centralised coordination, organisations risk inconsistent compliance and exposure to regulatory sanctions. The DPO serves as the 'central figure' for harmonising these requirements across legal, technical, and operational domains. 2️⃣Accountability and Demonstrable Compliance Supervisory authorities increasingly demand evidence-based compliance. Organisations must maintain detailed records of data flows, AI development processes, and incident responses. The DPO must champion a culture of #accountability, supported by robust governance structures and documentation protocols. 3️⃣ Technical and Organisational Complexity DORA mandates rigorous digital resilience testing and ICT risk assessments. The AI Act imposes strict data quality, explainability, and human oversight requirements. These obligations require cross-functional collaboration and significant investment in infrastructure, training, and tooling. At the end of the day, the DPO must act as a change agent, fostering alignment between compliance, innovation, and business objectives. The challenge is formidable, but so is the opportunity to redefine the role as a cornerstone of ethical, secure, and forward-looking digital governance.

If you want to be on top of Claude Code in 2026, you should know these 3 things clearly: 𝗦𝗸𝗶𝗹𝗹𝘀 𝘃𝘀 𝗦𝘂𝗯𝗮𝗴𝗲𝗻𝘁𝘀 𝘃𝘀 𝗣𝗹𝘂𝗴𝗶𝗻𝘀 — and when to reach for which. Three building blocks. Three very different jobs. Most teams use them interchangeably — and end up with agent systems harder to maintain than the code they replaced. Here's the simplest way I've found to decide: 𝗦𝗸𝗶𝗹𝗹𝘀 → reusable expertise → A folder with SKILL. md plus scripts, templates, references → Claude discovers and loads them when the task matches → Progressive disclosure: only relevant pieces enter context → Reach for skills when you want Claude to know how to do X consistently Use case: PDF generation, brand-compliant docs, domain-specific workflows 𝗦𝘂𝗯𝗮𝗴𝗲𝗻𝘁𝘀 → delegated specialists → A separate Claude instance with its own context, system prompt, tools → Main agent hands off, specialist returns a result → Keeps your primary context clean → Reach for subagents when the task is isolated and context-heavy Use case: Code review, deep research, parallel investigation, security audits 𝗣𝗹𝘂𝗴𝗶𝗻𝘀 → distributable bundles → A package combining skills, subagents, hooks, slash commands, MCP servers → Install once, entire capability stack arrives → Version-controlled, marketplace-ready → Reach for plugins when you're shipping to others or standardizing across teams Use case: Internal dev platforms, shared team toolchains, public releases The mental model: Skill = teach Claude a recipe Subagent = hire a specialist Plugin = ship the whole kitchen Most agent architectures fail not because the primitives are wrong — but because teams pick the wrong one for the job. A skill pretending to be a subagent leaks context. A subagent pretending to be a skill wastes tokens. A plugin without clear boundaries becomes a dependency nightmare. Start with the smallest unit that solves the problem. Compose upward only when you need to. What's your mental model for deciding between these three?

🛠️ If You Don’t Do This, Your CV Will Work Against You A few months ago, a Maintenance Manager at an offshore facility told me something wild: “I delete CVs with long summaries. If you need three paragraphs to tell me who you are, you’re not sure yourself.” And honestly? He’s right. Most CVs in oil & gas fail because they try too hard. Here’s what he said makes him stop reading: 📌 Long summaries “I am a passionate, hardworking, dedicated…” Bro, everyone is passionate when job hunting. 📌 Buzzwords If you write “team player” but can’t show a project where you actually worked with a team? It’s noise. 📌 Listing every software on earth Petrel. AutoCAD. SolidWorks. Excel. MATLAB. Python. PLCs. All on the same CV. (You’ll confuse the ATS and the hiring manager.) 📌 20 responsibilities with no outcome “Carried out maintenance on pumps.” “Assisted in inspection.” “Worked with field team.” This tells them NOTHING about your actual impact. → What to Do Instead (and Why It Works) Imagine two technicians: Tech A: Writes: “Carried out pump maintenance.” Tech B: Writes: “Reduced pump vibration by 18% after realigning coupling on Sulzer pump using dial indicators.” Guess who gets the interview? The second one because they proved competence. So use this instead: 📌 3 Key Achievements Pick things that made operations safer, faster, or cheaper. Example: • Cut troubleshooting time by 30 minutes using improved fault isolation. • Improved greasing routine for rotating equipment reduced failures. 📌 1 Small Project You Built This is your credibility booster. Example: • A quick report I created on common separator failures and simple field fixes. 📌 Tools/Equipment You Used This helps managers imagine you on their site. Examples: • Dial indicators • Multimeter • Calibration pump • Centrifugal pumps • Compressors • PLC panels 📌 1 Clear Specialization Instead of “Mechanical Technician | HSE | Electrical, try: “Mechanical Technician Pumps • Compressors • Rotating Equipment” That level of clarity gets callbacks. 📌 A Short Value Proposition One line that tells them what you bring: “I improve equipment reliability by identifying and correcting small inefficiencies before they become downtime.” That’s how you show value without bragging. The Truth: → Your CV should not tell your life story. → It should show the specific engineering value you bring to a site, crew, or operation. 👉🏼 Your CV should answer one silent question from every manager: ‘Can this person make my job easier? Make your CV a tool, not a biography. #CareerTips #EngineeringCV #OilAndGas #JobSearch #EnergyIndustry

Using Claude without a system is where most people lose time. Same tool. Completely different results - depending on how you use it. This model tree breaks down how to choose the right Claude model and workflow based on your task. Start simple: → Quick tasks, short prompts → Sonnet 4.6 → Lightweight work, saving tokens → Haiku 4.5 When things get complex: → Deep reasoning, multi-step work, file-heavy tasks → Opus 4.7 But the real value is not just model selection. It’s how you structure your workflow. → Plan first, then execute → Ask clarifying questions before jumping into outputs → Batch tasks instead of sending multiple messages → Convert files into structured formats for better results → Use extended thinking when depth matters And one underrated habit: Token efficiency. Small changes like editing prompts, combining tasks, and controlling output format can dramatically improve quality and cost. Also: → Use connectors (Drive, Slack, Notion) for real workflows → Upload structured files instead of raw PDFs → Save sessions and reuse workflows This is the shift: From chatting with AI → to operating AI like a system Once you follow a structured flow, Claude becomes predictable, faster, and far more powerful.

𝗜𝗳 𝘆𝗼𝘂'𝗿𝗲 𝘂𝘀𝗶𝗻𝗴 𝗖𝗹𝗮𝘂𝗱𝗲 𝗖𝗼𝗱𝗲 𝘄𝗶𝘁𝗵 𝗱𝗲𝗳𝗮𝘂𝗹𝘁 𝘀𝗲𝘁𝘁𝗶𝗻𝗴𝘀, 𝘆𝗼𝘂'𝗿𝗲 𝗹𝗲𝗮𝘃𝗶𝗻𝗴 𝟭𝟬𝘅 𝗼𝗻 𝘁𝗵𝗲 𝘁𝗮𝗯𝗹𝗲. Boris Cherny (creator of Claude Code) just shared how the Claude Code team actually uses it to get 10x productivity. 𝟭) 𝗥𝘂𝗻 𝟯-𝟱 𝘀𝗲𝘀𝘀𝗶𝗼𝗻𝘀 𝗶𝗻 𝗽𝗮𝗿𝗮𝗹𝗹𝗲𝗹 → Use Git worktrees to run parallel Claude Code sessions on the same repo so you can progress multiple threads without mixing context. 𝟮) 𝗦𝘁𝗮𝗿𝘁 𝗰𝗼𝗺𝗽𝗹𝗲𝘅 𝘁𝗮𝘀𝗸𝘀 𝗶𝗻 𝗽𝗹𝗮𝗻 𝗺𝗼𝗱𝗲 → For complex tasks, start in Plan Mode so Claude drafts the approach, you review it, and you re-plan early instead of debugging chaos later. 𝟯) 𝗧𝗿𝗲𝗮𝘁 𝗖𝗟𝗔𝗨𝗗𝗘.𝗺𝗱 𝗹𝗶𝗸𝗲 𝗮 𝗺𝗲𝗺𝗼𝗿𝘆 𝘀𝘆𝘀𝘁𝗲𝗺 → Treat CLAUDE.md like a memory system by updating it after every correction so the same mistake stops happening over time. 𝟰) 𝗧𝘂𝗿𝗻 𝗿𝗲𝗽𝗲𝗮𝘁 𝘄𝗼𝗿𝗸 𝗶𝗻𝘁𝗼 𝘀𝗸𝗶𝗹𝗹𝘀 → Convert anything you do repeatedly into a Skill so your workflows, checks, and commands become reusable instead of re-explained each time. 𝟱) 𝗟𝗲𝘁 𝗖𝗹𝗮𝘂𝗱𝗲 𝗳𝗶𝘅 𝗯𝘂𝗴𝘀 𝗮𝘂𝘁𝗼𝗻𝗼𝗺𝗼𝘂𝘀𝗹𝘆 → Fix bugs autonomously by pasting the bug thread plus failing tests and logs, then having Claude run the full diagnose-to-fix loop. 𝟲) 𝗨𝘀𝗲 𝗖𝗹𝗮𝘂𝗱𝗲 𝗮𝘀 𝗮 𝗵𝗮𝗿𝘀𝗵 𝗿𝗲𝘃𝗶𝗲𝘄𝗲𝗿 → Use a harsh reviewer pass where Claude challenges your changes to prove they work and pushes you toward the elegant implementation: - “Grill me on these changes and don’t open a PR until I pass” - “Prove this works” (diff main vs feature branch) - After a weak solution: “Scrap this and implement the elegant version” 𝟳) 𝗧𝗲𝗿𝗺𝗶𝗻𝗮𝗹 𝘀𝗲𝘁𝘂𝗽 𝗺𝗮𝘁𝘁𝗲𝗿𝘀 → Invest in terminal setup because fast feedback, clear status, and low-friction tooling matter more than fancy prompting. 𝟴) 𝗨𝘀𝗲 𝘀𝘂𝗯𝗮𝗴𝗲𝗻𝘁𝘀 → Use subagents to offload narrow tasks so your main Claude session stays clean, focused, and high-context. 𝟵) 𝗖𝗹𝗮𝘂𝗱𝗲 𝗳𝗼𝗿 𝗮𝗻𝗮𝗹𝘆𝘁𝗶𝗰𝘀 → Use Claude for analytics by turning SQL and data queries into a repeatable skill layer that ships decisions faster. 𝟭𝟬) 𝗟𝗲𝗮𝗿𝗻 𝘄𝗶𝘁𝗵 𝗖𝗹𝗮𝘂𝗱𝗲 → Learn with Claude by explaining your understanding so it fills gaps, stores the result, and reinforces it with spaced repetition. Claude Code is becoming a platform you shape around how you work. 𝗣.𝗦. 𝗜 𝘄𝗿𝗶𝘁𝗲 𝗮 𝗳𝗿𝗲𝗲 𝘄𝗲𝗲𝗸𝗹𝘆 𝗻𝗲𝘄𝘀𝗹𝗲𝘁𝘁𝗲𝗿 𝗼𝗻 𝗔𝗜 𝗮𝗴𝗲𝗻𝘁𝘀, 𝘄𝗼𝗿𝗸𝗳𝗹𝗼𝘄𝘀, 𝗮𝗻𝗱 𝘁𝗵𝗲 𝘀𝗸𝗶𝗹𝗹𝘀 𝗽���𝗼𝗳𝗲𝘀𝘀𝗶𝗼𝗻𝗮𝗹𝘀 𝗻𝗲𝗲𝗱 𝘁𝗼 𝘀𝘁𝗮𝘆 𝗮𝗵𝗲𝗮𝗱: https://lnkd.in/dbf74Y9E

OpenAI, Anthropic & Google engineers don't write prompts like you. They do this one thing you don't: Context engineering (more than just prompting). I explained everything here: https://lnkd.in/d7gBvDmK. But here's a side-by-side comparison: ❌ Bad prompt: "Act like a senior strategist." ✅ Good prompt: "Act like a senior strategist. You're advising a 50-person SaaS company. The CEO cares about speed-to-market, not perfection." ❌ Bad prompt: "Make it accurate." ✅ Good prompt: "The facts should be verifiable - something a reader could fact-check in 2 minutes online using this website [link]." ❌ Bad prompt: "First, analyze the data. Then, identify patterns. Finally, write conclusions." ✅ Good prompt: "I need an analysis where stakeholders can understand the key patterns without a technical background." ❌ Bad prompt: "Follow these rules: 1) Keep it brief, 2) Use simple language, 3) No jargon." ✅ Good prompt: "This is for someone encountering this topic for the first time, so focus on clarity and practical examples." ❌ Bad prompt: "Be concise." ✅ Good prompt: "Each section should fit in a single paragraph so this works as a quick reference." ❌ Bad prompt: "Explain it well." ✅ Good prompt: "Someone should understand this without searching for definitions of technical terms." ❌ Bad prompt: "Write me a strategy document." ✅ Good prompt: "I need a strategy document that helps my team decide whether to adopt an AI tool. This is for non-technical product managers. Success means they can explain 3 key decisions to leadership, not that they understand the underlying architecture." Stop telling AI what to do. Start telling AI what success looks like. Stick to these 5 principles while prompting: 1. Goals > Instructions Tell AI what you're trying to achieve, not how to achieve it. 2. Constraints > Rules Specify boundaries and priorities. Let the model choose execution. 3. Examples > Drescriptions Show what you want instead of describing it. 4. Put important info at the start & end Not in the middle (AI literally forgets the middle). 5. Less is more Every extra token exhausts AI's attention. Prompt engineering: "What you say." Context engineering: "Everything else the AI sees." That includes: → Examples → Memory → Output format → Tools → Who the audience is → What failure looks like Send it to a colleague who's (wrongly) obsessed with prompt engineering. It's all about context.

Most freshers entering Cyber Security make one common mistake: They try to learn “everything” instead of learning the tools actually used in real SOC environments. So I created this simple roadmap of the most important tools every: • Fresher • SOC Analyst aspirant • Career switcher into SOC should learn to become more job-ready for real-time Security Operations Center roles. The focus should not only be on certifications. The real goal is understanding how analysts actually: ✔ Investigate alerts ✔ Analyze logs ✔ Handle incidents ✔ Detect threats ✔ Respond to attacks Some of the most important categories include: 🔹 SIEM Tools 🔹 Endpoint Security / EDR 🔹 Identity & Access Management 🔹 Threat Intelligence 🔹 Networking & Monitoring 🔹 SOAR & Automation 🔹 Cloud Security 🔹 Linux & Windows Fundamentals Tools like: • Splunk • Microsoft Sentinel • Microsoft Defender for Endpoint • Wireshark • Microsoft Entra ID • CrowdStrike Falcon are highly valuable in today’s SOC ecosystem. If you are starting your journey: Start with fundamentals first. Then move into SIEM + EDR + Incident Investigation. That combination alone can make you stand out for many SOC L1 opportunities. Consistency > learning too many tools at once. Which SOC tool are you currently learning? 👇 #CyberSecurity #SOCAnalyst #SIEM #EDR #ThreatHunting #BlueTeam #CyberSecurityJobs #Splunk #MicrosoftSentinel #Defender #SOC #CareerSwitch #Freshers #InformationSecurity #CyberDefense #Learning #TechCareer

To perform their duties responsibly, boards must function as Humans + AI. Adopting new working structures and evolved governance structures incorporating AI can lead to substantial performance improvement. Much of my current work with boards is on strategic framing for AI and in AI-augmented decision-making, but there is considerably more potential. A very nice HBR piece brings real-world insights to bear. The first finding was that directors and chairs largely failed to recognize the value and potential of AI in their work. However still many boards and directors are using AI in useful ways. MEETING PREPARATION Directors who use LLMs reported significantly improved understanding of agenda items and reduced workload. One director across five Danish boards uses AI to structure presentations and run simulations; another in Switzerland uses it to refine board discussion questions from the board book. SCENARIO PLANNING GenAI, used well, can be an excellent tool for rapid scenario planning. One board in Austria used an LLM to analyze geopolitical risk in an acquisition proposal. This led to it rejecting the deal, and resulted in management attaching scenario analyses to future proposals. ADDITIONAL PERSPECTIVES Boards in Finland and the Netherlands used AI to test their own strategic conclusions, finding significant overlap between AI-generated insights and their human decisions. This boosted both their confidence in the decisions and their trust in AI’s utility, particularly for validating or challenging complex judgments. IMPROVING BOARD DYNAMICS AI can offer real-time feedback on boardroom dynamics. For example, a Swiss industrial company uses AI to analyze speaking time, tone, and engagement during meetings, creating recommendations for better group engagement. The article addresses potential risks: 🔐 Information leaks. These stem not from AI itself but from poor data governance, which can be mitigated with proper access controls and security training. ⚖️ Sample bias. Regular audits and user awareness are key to avoiding flawed, discriminatory, or incomplete insights. 🧭 Anchoring in the past. AI can be overly reliant on historical data. Scenario simulations and reasoning models can help boards anticipate and adapt to future shifts. And concludes with recommendations on learning to use AI well: 1️⃣ Create engagement. Chairs should start with one-on-one conversations to assess AI literacy and follow up with tailored training to build confidence and interest. 2️⃣ Practice collective experimentation. Boards should test AI tools together in low-stakes settings, debrief their experiences, and gradually integrate AI into governance processes. 3️⃣ Maintain momentum. Chairs must lead by example, celebrate AI use regardless of outcomes, and embed AI progress into board evaluations. I am currently working on a 'GenAI in the Boardroom' mini-report that I will be sharing soon, addressing these and a range of other issues and possibilities.