# Ayan Sarkar | The Practical CTO

> The Practical CTO is the personal publication of Ayan Sarkar, Chief Technology & AI Officer and Co-Founder of Webskitters. It offers practical perspectives for founders and technology leaders on enterprise AI strategy, agentic systems, private LLMs, technology architecture, modernisation and technology leadership.

Written in UK English. Articles are analysis and opinion drawn from operating experience; they are not legal, financial or investment advice. Each article and page is also available as Markdown: append .md to its URL without the trailing slash, for example https://thepracticalcto.com/blog/private-llm-for-enterprise.md.

## About Ayan
- [About](https://thepracticalcto.com/about/): Background, role at Webskitters and the perspective behind this publication
- [Speaking](https://thepracticalcto.com/speaking/): Topics, formats and speaking enquiries
- [Contact](https://thepracticalcto.com/contact/): Ways to engage: strategic advisory, architecture and AI reviews, executive workshops, speaking and media

## Expertise
- [Expertise overview](https://thepracticalcto.com/expertise/): The disciplines Ayan works across, from AI strategy to architecture and modernisation
- [Enterprise AI strategy](https://thepracticalcto.com/expertise/enterprise-ai-strategy/): Where AI should be used, how to prioritise it and how to move from pilots to capability
- [Agentic AI for enterprise](https://thepracticalcto.com/expertise/agentic-ai-for-enterprise/): Designing, governing and operating AI agents that take actions
- [Private LLMs and enterprise knowledge](https://thepracticalcto.com/expertise/private-llms-enterprise-knowledge/): When private deployment is justified and why the knowledge layer matters more

## Advisory services
- [Services overview](https://thepracticalcto.com/services/): Technology and AI advisory, organised by the situation a leadership team is in
- [Fractional CTO](https://thepracticalcto.com/services/fractional-cto/): A Fractional CTO provides embedded executive technology leadership on a part-time basis, taking responsibility for technology strategy, architecture, engineering capability, AI adoption, delivery and technology risk without requiring a full-time CTO appointment. Unlike an adviser, a Fractional CTO is accountable for outcomes: they sit with the leadership team, lead the engineering organisation and own technology decisions until a permanent leader is needed or appointed.
- [Virtual CTO](https://thepracticalcto.com/services/virtual-cto/): A company should use a Virtual CTO when it faces technology decisions that need senior, independent judgement but does not need an executive to lead the engineering team. Typical moments are choosing a platform or development partner, validating an architecture, deciding on an AI approach, or reviewing a roadmap before committing budget. The Virtual CTO advises and reviews; the decision and the day-to-day management stay with you.
- [AI Readiness Assessment](https://thepracticalcto.com/services/ai-readiness-assessment/): An AI readiness assessment evaluates whether an organisation has the strategy, data, architecture, people, governance and operational capability required to deploy AI responsibly and generate measurable business value. It identifies which gaps would cause AI initiatives to stall or create risk, and sets out the order in which to close them, so investment goes into use cases the organisation can actually run.
- [AI Strategy Advisory](https://thepracticalcto.com/services/ai-strategy-advisory/): Enterprise AI strategy advisory helps a leadership team decide where AI should be used, in what order, with which architecture and under what governance, so that AI investment produces measurable business value rather than disconnected pilots. It turns a broad ambition into a prioritised set of use cases, a build-or-buy position, an operating model and a roadmap the organisation can actually deliver.
- [Technology Architecture Review](https://thepracticalcto.com/services/technology-architecture-review/): A technology architecture review is an independent assessment of how a platform is designed, built and operated, and whether it can support the organisation’s plans. It examines application structure, infrastructure, data, integration, security, reliability, technical debt and cost, then separates the risks that matter commercially from those that do not, and sets out a prioritised plan for modernisation.
- [AI Architecture Review](https://thepracticalcto.com/services/ai-architecture-review/): An AI architecture review covers how an AI system uses models, knowledge, tools and people, and whether it is accurate, secure, governable and affordable in production. It examines model choice and routing, retrieval and knowledge quality, agent permissions and tool use, human approval, evaluations, observability, security, auditability, cost and latency, and identifies which weaknesses carry real business risk.
- [Technical Due Diligence](https://thepracticalcto.com/services/technical-due-diligence/): Technical due diligence is an independent assessment of a company’s technology, engineering organisation and technical risks, carried out to inform an investment, acquisition or major commercial decision. It tests whether the technology can support the business plan, what it will cost to get there, and whether claims about the product, including AI capabilities, stand up to evidence.
- [Executive AI Workshops](https://thepracticalcto.com/services/executive-ai-workshops/): An executive AI workshop is a facilitated working session that helps a leadership or engineering team build a shared understanding of what AI can and cannot do for their organisation, and use it to make decisions. Unlike a talk, it is tailored to the organisation’s context and ends with agreed priorities, open questions and next steps.

## Answers and frameworks
- [Ask Ayan](https://thepracticalcto.com/ask-ayan-ai/): Curated, source-linked answers to common technology and AI questions, written by Ayan. Not a chatbot
- [AI Lab](https://thepracticalcto.com/ai-lab/): Frameworks and design notes for applying AI in organisations
- [Newsletter archive](https://thepracticalcto.com/thinking/): Editions of The Practical CTO newsletter, first published on LinkedIn

## Flagship insights
- [From Pilots to Capability: Why Enterprise AI Stalls](https://thepracticalcto.com/blog/from-pilots-to-capability-why-enterprise-ai-stalls/): The pilots work; the business does not change. Why enterprise AI stalls between the demo and the operating model — and what building real capability involves.
- [Private LLM for Enterprise: When It Is Worth It](https://thepracticalcto.com/blog/private-llm-for-enterprise/): Private LLM for enterprise, honestly assessed: the four deployment models, the guarantees each one actually buys, the total cost people underestimate, and a decision checklist.
- [An AI Governance Framework CTOs Can Actually Run](https://thepracticalcto.com/blog/ai-governance-framework/): A practical AI governance framework for CTOs: four operating layers — accountability, risk, controls and evidence — mapped to NIST AI RMF and the EU AI Act, with a cadence you can actually run.
- [CTO vs CIO vs Chief AI Officer: Who Owns AI?](https://thepracticalcto.com/blog/cto-vs-cio-vs-chief-ai-officer/): CTO vs CIO vs Chief AI Officer — a practical look at what each role owns, why AI cuts across all three, and how to decide who should be accountable in your organisation.
- [What AI Changes About Engineering Culture — and What It Should Not](https://thepracticalcto.com/blog/what-ai-changes-about-engineering-culture/): AI-assisted development is a genuine shift in how software gets built. What leaders should embrace, what they should protect — and why judgment and ownership matter more, not less.
- [Enterprise Knowledge Is the Real AI Advantage](https://thepracticalcto.com/blog/enterprise-knowledge-is-the-real-ai-advantage/): Models are available to everyone; your organisation's knowledge is not. Why the condition of enterprise knowledge — ownership, access, quality — decides what AI can actually do for you.
- [Designing for Agents: When Software Starts to Act](https://thepracticalcto.com/blog/designing-for-agents-when-software-starts-to-act/): Agents move software from answers to actions. The design questions that matter now are about permissions, evidence of completion, recovery and the limits of delegation.

## Latest analysis
- [Multi Agent Systems for reliable QA agents](https://thepracticalcto.com/blog/multi-agent-systems-agentic-qa-blueprint/): Build agentic QA with Multi Agent Systems: wire LLMs to browser automation, minimise failing tests, capture state, and de-flake with golden traces.
- [AI-powered cyberattacks: a 90‑day CTO plan](https://thepracticalcto.com/blog/ai-powered-cyberattacks-90-day-cto-plan/): A 90‑day CTO plan to prepare for AI-powered cyberattacks: identity, segmentation, AI-aware detection, code-signing, and IR rehearsals.
- [Marvell Lumentum merger: de-risk 1.6T](https://thepracticalcto.com/blog/marvell-lumentum-merger-de-risk-800g-1-6t/): The Marvell Lumentum merger tightens AI connectivity supply. Here’s a practical plan to de-risk 800G/1.6T rollouts with multi-sourcing and interop tests.
- [AI Strategy for Enterprises: Siting Risk](https://thepracticalcto.com/blog/ai-strategy-for-enterprises-siting-risk/): AI Strategy for Enterprises now includes siting risk. Add energy, water and carbon SLOs, curtailment clauses and fallback capacity to keep roadmaps on…
- [AI-assisted code governance, the Linux way](https://thepracticalcto.com/blog/ai-assisted-code-governance-linux-template/): A 30‑day AI-assisted code governance rollout: disclose AI in commits, capture provenance, tighten reviews, and block bulk patch drops.
- [Samsung advanced packaging: your 2027 plan](https://thepracticalcto.com/blog/samsung-advanced-packaging-2027-ai-hardware-plan/): Samsung advanced packaging focus in Japan signals supply moves. Here’s how to rebalance 2027 AI hardware: dual-qual, pre-book capacity, extend buffers.
- [Enterprise AI Agents: LLM‑aware defences](https://thepracticalcto.com/blog/enterprise-ai-agents-llm-aware-defences/): Malware now uses LLMs to triage victims. Ship LLM‑aware defences for Enterprise AI Agents: egress allowlists, model‑load blocks, and canary traps.
- [Claude Code restrictions: hedge your delivery](https://thepracticalcto.com/blog/claude-code-restrictions-hedge-your-delivery/): Claude Code restrictions broke reverse‑engineering tasks. Here’s how to audit, set compliant boundaries, and add fallbacks so safety shifts don’t stall…
- [Private LLM for Enterprise: Mistral Large 4](https://thepracticalcto.com/blog/private-llm-for-enterprise-mistral-large-4/): A 48‑hour, practical playbook to evaluate and self‑host Mistral Large 4 as a Private LLM for Enterprise: VRAM targets, quantisation, serving, and risks.
- [OpenAI funding target cut: your next moves](https://thepracticalcto.com/blog/openai-funding-target-cut-capacity-pricing-plan/): OpenAI funding target cut: what it means for capacity, pricing and your 2025–27 AI plan, plus a vendor, procurement and architecture checklist.
- [AI Spending Backlash: Your 2025 Contract Play](https://thepracticalcto.com/blog/ai-spending-backlash-2025-cloud-ai-contracts/): AI spending jitters have hit Big Tech. Here’s how to use the pressure to reshape 2025 cloud and AI contracts, pricing and risk on your terms.
- [Enterprise AI Strategy: Weighing Gemini 4 Carbon](https://thepracticalcto.com/blog/enterprise-ai-strategy-gemini-4-carbon-decision-memo/): CIO decision memo on Gemini 4 Carbon vs OpenAI/Mistral: security, latency, TCO and data residency, plus a 30‑day rollout plan. Enterprise AI Strategy.

## Optional
- [All Insights](https://thepracticalcto.com/blog/): Every article, newest first
- [Full text of all articles](https://thepracticalcto.com/llms-full.txt)
- [RSS feed](https://thepracticalcto.com/feed/)
- [XML sitemap](https://thepracticalcto.com/sitemap.xml)