RichardHam.co.uk Blog

Tag: AI

  • How Multi-Agent Operations Work in Practice

    How Multi-Agent Operations Work in Practice

    Most organisations still treat AI agents like a single chat window with extra buttons. That is fine for a demo. It is not fine when the work touches production systems.

    The difference is operational, not magical. The teams getting value from agents are the ones that add roles, checkpoints, and ownership.

    What works

    A reliable setup usually has four pieces:

    • a coordinator that defines the task and checks the output
    • worker agents that do independent chunks in parallel
    • state that lives somewhere everyone can inspect
    • review points before anything risky moves forward

    That is not glamorous, but it works.

    A simple example

    If three services are acting up at once, a good coordinator breaks the problem apart and sends each service to a separate worker. The workers do not need to talk to each other because the tasks are independent. The coordinator then compares the results and decides whether to approve the fix or escalate.

    That pattern saves time without turning the system into a black box.

    The cost question

    People often assume agent work must be expensive. In practice, the opposite is usually true. The cheap model can do the repetitive work. The better reasoning model is reserved for coordination and review.

    That split matters. It keeps the system affordable and keeps judgment where it belongs.

    What to do first

    1. Decide who owns the outcome.
    2. Identify which tasks can run in parallel.
    3. Make state visible.
    4. Add checkpoints where mistakes would hurt.

    That is enough to get started. The rest is tuning.

  • How Multi-Agent Operations Work in Practice

    e

    Most organisations still treat AI agents like a single chat window with extra buttons. That is fine for a demo. It is not fine when the work touches production systems.

    The difference is operational, not magical. The teams getting value from agents are the ones that add roles, checkpoints, and ownership.

    What works

    A reliable setup usually has four pieces:

    • a coordinator that defines the task and checks the output
    • worker agents that do independent chunks in parallel
    • state that lives somewhere everyone can inspect
    • review points before anything risky moves forward

    That is not glamorous, but it works.

    A simple example

    If three services are acting up at once, a good coordinator breaks the problem apart and sends each service to a separate worker. The workers do not need to talk to each other because the tasks are independent. The coordinator then compares the results and decides whether to approve the fix or escalate.

    That pattern saves time without turning the system into a black box.

    The cost question

    People often assume agent work must be expensive. In practice, the opposite is usually true. The cheap model can do the repetitive work. The better reasoning model is reserved for coordination and review.

    That split matters. It keeps the system affordable and keeps judgment where it belongs.

    What to do first

    1. Decide who owns the outcome.
    2. Identify which tasks can run in parallel.
    3. Make state visible.
    4. Add checkpoints where mistakes would hurt.

    That is enough to get started. The rest is tuning.

  • HeliOS-Studio: AI Startup Studio Ignites

    t

    After a few years of building AI tooling, I hit a point where the stack stopped feeling like scaffolding. It started to look like a product in its own right.

    That is a strange moment. You begin by solving a narrow operational problem, then realise the workflow you built to support the work is now valuable enough to stand on its own.

    The shape of it

    The setup was simple in principle:

    • one layer for orchestration
    • one layer for safe execution
    • one layer for inference
    • one layer for content and delivery

    The names changed over time. The pattern did not.

    What the studio produced

    The useful output was not a single breakthrough. It was a steady stream of small, shippable things: business plans, MVP outlines, content drafts, docs, and working repos.

    That changed how I thought about progress. Instead of asking, “Can the system automate this?” I started asking, “Can the system help turn this into something a person could actually use?”

    The takeaway

    Infrastructure is only boring until it starts making decisions for you.

    When the workflow is good enough, the tooling stops being background noise. It becomes part of the offer.

  • EU AI Act Compliance: Governance Frameworks in Practice

    EU AI Act: My Clients Were Ready. Most Weren’t.
    Published: November 10, 2025 (retrospective)

    EU AI Act enforcement began in earnest in late 2025. While many businesses scrambled, my clients had zero compliance findings across seven audits. The governance habits built into SentinelForge since 2024—audit trails, human gates, scoped permissions—turned out to be exactly what regulators wanted to see.

    Framework Coverage

    Framework Status Coverage Area
    EU AI Act ✅ Complete High-risk AI systems
    NIST AI RMF ✅ Complete Full stack governance
    ISO 42001 80% Audit-ready
    OECD AI Principles ✅ Complete Transparency + accountability

    What Auditors Actually Look For

    1. Audit trail completeness — every AI decision logged with timestamp and rationale
    2. Human oversight documentation — evidence that humans reviewed high-risk outputs
    3. Data governance — proof that personal data wasn’t used to train models without consent

    SentinelForge’s GitHub-gated architecture satisfied all three out of the box. The logs were already there.

    The Lesson

    Compliance isn’t a bolt-on. The businesses that struggled in 2025 were those that treated AI governance as a 2025 problem. We started in 2023.

    Need EU AI Act readiness for your AI systems? Book a governance audit.

    Next: HeliOS-Studio—AI startup studio ignites (Feb 2026).

  • AI Arms Race: Predictive Cyber Defence

    AI Arms Race: Predictive Cyber Defence Is Here
    Published: August 20, 2025 (retrospective)

    The AI cybersecurity market is projected to hit $60B by 2028—and for good reason. In August 2025, SentinelForge v2’s predictive threat hunting caught a client ransomware pivot 72 hours before it would have detonated. No SOC. No SIEM subscription. Just CrewAI agents, local LLMs, and disciplined governance.

    SentinelForge v2 Production Stack

    proxmox-ve
└── sentinelforge (docker)
    ├── crewai crews     (24/7 autonomous monitoring)
    ├── ollama           (local inference)
    ├── grafana          (observability)
    └── uptimekuma       (SLA: 99.9%)

    The Catch: Anatomy of a Prevention

    • Day 1: Anomalous LDAP query pattern flagged by Audit Crew
    • Day 2: Lateral movement indicators correlated across 3 systems
    • Day 3 (72h): Human review triggered; client isolated affected segment
    • Result: Zero encryption, zero ransom, zero downtime

    What This Means for SMEs

    Enterprise-grade predictive defence is now accessible without enterprise budgets. The stack cost: £0/month in cloud tokens, running on repurposed hardware.

    1. AI agents don’t get tired—24/7 monitoring without alert fatigue.
    2. Local inference keeps sensitive threat data off third-party servers.
    3. Governance logs every detection decision—invaluable for insurance and compliance.

    Want predictive AI defence for your business? Book a Secure AI QuickScan.

    Next: EU AI Act compliance—governance frameworks in practice (Nov 2025).

  • What AI Session Logs Can Tell You About How You Work

    s

    I had spent a long time using AI tools for code, ideas, and problem-solving. At some point I started wondering whether the logs might be useful for something other than debugging the tools.

    So I exported a pile of sessions and looked for patterns.

    The pipeline

    chat exports -> normalise -> analyse -> notes

    The point was not to turn private data into a product. The point was to understand my own habits without relying on memory or vibes.

    What turned up

    1. I over-engineer security more often than I notice in the moment.
    2. I default to doing things myself even when delegation would save time.
    3. I think in systems now, not isolated tasks.

    None of that was shocking, but seeing it written down made it harder to ignore.

    Why it mattered

    The useful part was not the novelty. It was the feedback loop. AI logs can show you where you repeat yourself, where you hesitate, and where you keep solving the same problem in slightly different ways.

    That can be useful. It can also be a privacy trap if you treat the data casually.

    My rule stayed simple: keep the raw data local, keep the analysis honest, and do not mistake pattern recognition for wisdom.

  • What 2024 Taught Me About Turning AI Work Into Infrastructure

    s

    2024 was the year AI stopped feeling like a side experiment and started feeling like part of the working stack.

    The biggest change was not speed. It was structure.

    What changed

    • routine work got easier to delegate
    • review became part of the workflow instead of an afterthought
    • local execution mattered more than hype
    • security thinking moved earlier in the process

    The numbers were useful, but the bigger shift was behavioural. I spent less time wrestling with one-off tasks and more time building repeatable paths.

    What worked

    The pattern that kept showing up was simple:

    1. keep the sensitive bits local when possible
    2. make the outputs reviewable
    3. use the repository as the record of truth

    That combination did more for consistency than any single tool choice.

    What I would change

    If I were doing it again, I would write more of the operating rules earlier. The systems worked better once the guardrails were explicit.

    That is usually how these things go. The tech is rarely the hard part. The hard part is deciding how much freedom the workflow should really have.

  • How I Stopped AI Workflows From Sprawling

    g

    By mid-2024 I was juggling several AI tools at once. The work was useful, but the coordination started to eat the time I was trying to save. Every extra prompt, every follow-up, every half-finished idea added friction.

    So I changed the shape of the work. Ideas became GitHub issues. Research happened first. Code changes went through review. Nothing moved forward without a human looking at it.

    The workflow

    Idea -> GitHub issue -> research -> code -> review -> deploy

    The important part was not the diagram. It was the discipline around it.

    • Every item carried a priority and a budget.
    • The review step stayed human.
    • The system ran overnight so I could review the output in the morning instead of reacting in real time.

    What it changed

    The main win was not speed for its own sake. It was clarity. I spent less time babysitting the process and more time making decisions that mattered.

    A rough summary of the difference:

    Metric Before After
    Code hours per week 20h 4h
    Monthly token spend £120 £12
    Projects shipped per month 1 4
    GitHub commits per month 45 200+

    What made it work

    1. Keep the routing local when you can.
    2. Keep write access scoped.
    3. Treat the repository as the source of truth.

    That combination is what made the setup durable. Not the tools, really. The rules.

    I still think that is the part people miss. AI does not remove the need for process. It makes process more obvious.

  • Why Zero-Trust Thinking Matters in AI Systems

    Zero-Trust + AI: Digital Transformation Gets Real
    Published: April 10, 2024 (retrospective)

    NIST’s AI Risk Management Framework (early 2024) collided with my Control Tower experiments. For years, I’d preached zero-trust to clients. Now I had to apply it to my own AI stack. Cybersecurity governance wasn’t optional anymore—SentinelForge planning began as a direct response.

    Governance Stack Emerges

    The principle was simple: every AI decision must be logged, auditable, and human-gated. The architecture:

    proxmox-ve
├── ollama          (local inference, no cloud leakage)
├── crewai          (agent orchestration, role-scoped)
├── vaultwarden     (secrets, zero plaintext)
└── github          (human approval gates on all PRs)

    This wasn’t theoretical. A March 2024 client incident—an AI-generated script with a subtle privilege escalation bug—proved every layer was necessary.

    Zero-Trust Applied to AI

    Principle Traditional IT AI Stack Application
    Verify explicitly MFA on every login Signed commits on every AI output
    Least privilege Minimal AD permissions Scoped agent tool access
    Assume breach EDR + SIEM Prompt injection detection

    Lessons

    1. Treat AI agents like privileged users—same controls, same audit trails.
    2. NIST AI RMF is practical, not theoretical; map it to your stack early.
    3. Digital transformation without governance is just technical debt with a faster delivery speed.

    Need a zero-trust AI framework for your business? Let’s talk.

    Next: Control Tower blueprints go live (Jul 2024).

  • M365 Copilot GA: Auditing in the AI Era

    t

    When Microsoft 365 Copilot arrived, the immediate question from clients was not “Can it help?” It was “Is it safe to turn on?”

    My answer was simple: not without a proper permissions review.

    AI does not remove tenant risk. It makes existing risk easier to surface.

    What I kept finding

    Across multiple audits, the same issues kept coming back:

    • over-permissive app consents
    • mailbox forwarding rules that nobody had reviewed in months
    • Intune drift that had crept in quietly

    None of that was exotic. It was just the usual gap between default settings and an actual security posture.

    Why the audit mattered

    The useful part of the process was not the reporting template. It was the discipline.

    I needed a way to check the tenant, explain the findings clearly, and keep the evidence somewhere traceable. That made the audit easier to repeat and a lot harder to hand-wave away.

    The lesson

    If Copilot is going to sit on top of the tenant, the tenant needs to be in decent shape first.

    That means permissions, logging, and a clear view of who can do what. Otherwise the tool just gives people faster access to a system they do not actually understand.

Powered by CookieAdmin