William Sandford
William SandfordOperations & AI EnablementBlenheim Chalcot

I build AI that closes the loop.

Practical agents that act on real business problems — not just flag them.

View selected work
01About

Operator building at the edge of AI.

Blenheim Chalcot took me on as an office assistant. I'm still here — office assistant became executive assistant, which folded into Operations & AI Enablement. Along the way I taught myself to vibe-code, started building AI tools for real business problems, and got Anthropic-certified to sharpen the skills properly. Self-taught and certified isn't a contradiction — it's probably how most useful skills get built now.

I've worked across the Blenheim Chalcot portfolio — HealthTech, MediaTech, GovTech at Agilisys — watching the same business problems turn up wearing different costumes. The problem is rarely the technology. More often it's the gap between the people building it and the people who need to trust and use it. One of those tools was for Babyzone, our early-years charity — I'm now training for the Battersea Park Marathon to raise money for them too.

I'm probably a better storyteller than I am a vibe-coder. Years in rooms with entrepreneurs who've spent decades building disruptive businesses teach you things a coding bootcamp doesn't — mainly, how to cut through the noise and tell a senior stakeholder the truth in a way they'll act on.

AI can write the code. It's much worse at reading the room. That's the gap I sit in.

02Selected work

Systems built around real problems.

Live tools, built and shipped for real teams — spanning GovTech, MediaTech, societal impact and internal operations.

ClickZ AI Agent Fleet

0kSubscriber operation run by 1 human + 10 agents
0ICP-scored leads surfaced per week

A fleet of ten named AI agents built on Paperclip, running ClickZ — Blenheim Chalcot's 40,000-subscriber MediaTech portfolio company — day to day with a single human overseeing it. At the centre sits Charlie: a coordination agent that can hire new agents, change an existing agent's job spec as priorities shift, and pull meeting transcripts to delegate the resulting tasks straight to people, not just to other agents. Below Charlie, agents cover the weekly newsletter, the flagship marketing-campaign column, LinkedIn lead generation, event coverage, video and independent security review — with humans still gating every send, publish and lead handoff.

10-agent fleet
Built on Paperclip
BoardHuman — ClickZ
CharlieCEO / Chief of StaffOPUS
RobCTO — Continuity
DougieCISO
IrisMeeting Intake
NoraNewsletter Editor
CleoCreatives EditorSONNET
EvanEvent Coverage
AriaAudience Growth
↳ reports to Aria
LennyLinkedIn AnalystSONNET
VioVideo ProducerSONNET

Charlie isn't fixed — it can hire a new agent, rewrite an existing agent's job spec as priorities shift, and pull meeting transcripts to delegate straight to people, not just to other agents.

Project 01MediaTech
ACTIVE
  • Paperclip
  • Claude Opus + Sonnet
  • Notion
  • HubSpot
  • GA4
  • Fathom
View case study

Daily Signal — Agilisys

£0mGovernment fund surfaced — now under bid
0+Public sources monitored daily

An automated daily intelligence digest for Agilisys, a Blenheim Chalcot GovTech portfolio company. Pulls live UK government tender, funding and policy data from 30+ public sources and uses AI to judge relevance and flag genuine business opportunities — no manual research required.

Pipeline running
Runs daily via GitHub Actions
  1. 1Sources
  2. 2Retrieve
  3. 3Filter
  4. 4Reason
  5. 5Brief
Live — Daily Signal — Agilisys
Project 02GovTech
ACTIVE
  • Claude API
  • GitHub Actions
  • Find a Tender / Contracts Finder
  • GOV.UK feeds
View case study

Daily Signal — Babyzone

A sister build of the Agilisys Daily Signal tool, adapted for Babyzone, Blenheim Chalcot's societal-impact portfolio company supporting under-5s in disadvantaged families. Tracks funding opportunities, grants, public-sector tenders and parliamentary activity relevant to the charity’s work, shaped directly through stakeholder interviews with its Fundraising and Policy teams.

Pipeline running
Runs daily via GitHub Actions
  1. 1Sources
  2. 2Retrieve
  3. 3Filter
  4. 4Reason
  5. 5Brief
Live — Daily Signal — Babyzone
Project 03Societal Impact
ACTIVE
  • Claude API
  • GitHub Actions
  • GOV.UK Find a Grant
  • Hansard
View case study

Autonomous Website Content Audit

A recurring audit of Blenheim Chalcot's and ClickZ's websites, built with Claude Code. Scores each site out of 100 across 50 metrics — from broken links and stale claims to how well content is optimised for AI/LLM search (GEO) — then suggests a specific action per metric and the score improvement it would deliver.

Audit running
Runs monthly
Site
About / People
Products
News / Careers
Legal / Footer
Score + Suggestions
Project 04Internal Tooling
ACTIVE
  • Claude Code
  • Parallel research agents
  • Scheduled runs
  • GEO / LLM scoring
  • MCP connectors
View case study

Portfolio Weekly Newsletter Agent

0Estimated hours saved per year
0Portfolio companies covered per run

A weekly internal newsletter agent for Blenheim Chalcot's portfolio of 9 companies. Each run scrapes live LinkedIn activity across every portfolio company, researches and fact-checks a GenAI news section, deduplicates against a rolling archive of prior editions, and produces a ready-to-review draft in three formats — without human input until the approval stage.

Pipeline running
Runs weekly via Claude Code scheduled task
  1. 1Scrape
  2. 2Research
  3. 3Dedupe
  4. 4Write
  5. 5Export ×3
Project 05Internal Tooling
ACTIVE
  • Claude Code
  • Claude in Chrome MCP
  • WebSearch / WebFetch
  • Pandoc
  • Scheduled tasks
View case study
03Experiments

Smaller, unfinished work — closer to a lab notebook than a product.

  • EXP-01Agent memory
    Instead of losing context across employees using Claude Cowork, using MCPs to keep context updated across multiple devices
  • EXP-02Browser agents
    Website development and HubSpot marketing changes using headless browser agents to do work that used to need a marketing agency
  • EXP-03AI voice
    Bringing one-man comedy shows into an immersive experience using AI voice
  • EXP-04Recursive agents
    Building organisations using Claude Code and Paperclip tools to build recursive, multi-agent workflows
  • EXP-05Human-in-the-loop
    Building approval checkpoints into agent workflows, guided by Anthropic's human-oversight principles
  • EXP-06Executive Command Centre
    Concept for an AI operating layer across inbox, calendar and travel — interface designed, not yet wired to live data
  • EXP-07Recruitment Intelligence
    Natural-language search prototype over candidate and meeting history — UI built, not yet connected to real records