Web Project Studios

Field notes

Honest writing about AI workflows in the wild.

Patterns, failure modes, and what actually works when you're shipping AI inside a small team. No content marketing.

Field notes

AI replaced the admin surface of recruiting. The job got harder.

A year ago, a high-profile prediction said AI would replace recruiters within six months. It was half right. Here is what actually got replaced, and what that means for every agency founder still standing.

13 July 2026

recruitmentai-workflowbusiness-development
Read more

13 Jul 2026

What the EU AI Act actually says about recruitment automation

Annex III of the EU AI Act classifies AI used in candidate sourcing and selection as high-risk. B2B business development is not. Agency founders need to draw that line now, before August 2026.

13 Jul 2026

Cold email for recruitment agencies without burning your domain

Most recruitment agencies run outbound email from their primary domain until something breaks. Here is how to build a sending infrastructure that survives contact with reality.

6 Jul 2026

The Apollo-to-Bullhorn shuffle is quietly eating your margin

Apollo dropped its native Bullhorn integration. Most agencies are filling the gap with copy-paste. Here is the playbook for replacing that with a durable integration pattern before the duplicate records and stale statuses compound.

6 Jul 2026

Deterministic prospect scoring beats AI guesswork for recruitment BD

Why your BD pipeline deserves a rubric calibrated against real close rates, not an LLM's confidence score. A worked scoring model any recruitment founder can audit.

6 Jul 2026

How to design an n8n workflow that degrades gracefully instead of failing silently

Most n8n workflows are built to succeed. This post introduces explicit degradation tiers so your automations handle partial failure without going dark.

29 Jun 2026

What a Rightmove listing quality score does not tell you about the agent who wrote it

Rightmove's listing quality indicators reward completeness and keyword density, not accuracy. An AI-generated description that invents a south-facing garden passes every automated check. That is the problem.

29 Jun 2026

Why your agency's AI content output all sounds like it came from the same writer

Marketing agencies running AI-assisted content at volume are producing work that converges on the same register. This is a workflow design problem, not a prompt quality problem.

29 Jun 2026

The Make scenario that runs every night and has never once been checked

Most Make automations reach production without alerting, log review cadence, or a named owner. 'It ran' is not the same as 'it worked.' Here is how to build observability in deliberately.

22 Jun 2026

What silent data corruption actually looks like in a production n8n workflow

The failure mode most workflow builders miss is not the loud crash. It is the field that maps wrong, the payload that truncates, the LLM node that returns a plausible but incorrect value. This post documents the patterns and the design constraints that prevent them.

22 Jun 2026

n8n error handling that works in testing silently fails in production

Most n8n error handling is tested against predictable failures. Production breaks differently. Here is why your error nodes stay quiet while bad data flows downstream.

22 Jun 2026

Why recruitment agency founders are still doing BD at 9pm

Most recruitment agency founders haven't got a BD problem. They've got a sequencing problem. The pipeline dries up exactly when billing peaks, and no hire fixes that.

15 Jun 2026

What the Renters' Rights Act does to your mid-tenancy communication workflow

The Renters' Rights Act didn't just abolish Section 21. It pulled out the fixed end date that your entire mid-tenancy workflow was built around. Here's what the replacement looks like.

15 Jun 2026

The client report that ships on time but says nothing useful

AI has made it trivially easy to generate a formatted monthly report in minutes. The problem is that speed has exposed what was always true: the report was never the deliverable, the interpretation was.

15 Jun 2026

The AML audit trail that looks complete until a solicitor actually requests it

Most estate agents believe a green SmartSearch or Credas result is their AML compliance. It is not. The obligation under Money Laundering Regulations 2017 is to document the decision, not just record the check.

10 Jun 2026

How to know whether your n8n workflow is actually running in production

Most n8n workflows have no alerting, no structured logging, and no health-check layer. They run fine until they silently stop. Here is how to build observability into automation before that happens to you.

8 Jun 2026

The agency intake form that collects everything and tells you nothing

Most marketing agencies have a thorough client intake form. The problem is not the form. It is the gap between submission and kickoff where nobody reads it.

8 Jun 2026

What a Make scenario looks like six months after the person who built it left

Automation debt is not a technical problem. It is an institutional memory problem. Here is what happens to a Make scenario when the builder leaves and nobody documents anything.

5 Jun 2026

Rightmove descriptions written by AI are converging on the same four adjectives

Most mid-size agencies switched to AI-assisted listing copy months ago. The output has homogenised. This is not a tool problem; it is a brief problem with a specific structural fix.

3 Jun 2026

The EU AI Act readiness check for small UK businesses

Most small UK agencies assume the EU AI Act is someone else's problem. Article 2 disagrees. Here is a one-page readiness check for 3-15 person teams using AI in their workflows.

1 Jun 2026

The retainer that looked profitable until you mapped the handoffs

Marketing agency retainers rarely go unprofitable in one moment. They drift through accumulated handoffs: extra amends rounds, unscoped coverage, reporting that takes three hours instead of one. The unit of analysis is not deliverable count but handoff count.

28 May 2026

Hiring AI under the EU AI Act: the deadline moved, the obligations did not

The EU pushed the Annex III high-risk deadline from August 2026 to December 2027. That changes the calendar, not the requirements. If your hiring AI has no risk management system, no audit log, and no human oversight mechanism, sixteen extra months is not a reprieve. It is a build window.

25 May 2026

The Renters' Rights Act does not care how good your template is

Lettings agents are updating tenancy agreement templates ahead of the Renters' Rights Act, but document quality is not process readiness. The real exposure is in the sequence: periodic transitions, rent increase records, and the Section 8 audit trail that county court will actually inspect.

18 May 2026

The lettings agent's tenant scoring blind spot under the EU AI Act

AI-augmented tenant referencing tools may already classify your agency as a deployer of a high-risk AI system under the EU AI Act. Most agents don't know. The obligations apply anyway.

18 May 2026

The lettings inspection report that won't survive the next case

Post-Renters' Rights Act, Section 8 possession cases turn on evidence quality, not whether the inspection happened. Most letting agents' reports won't hold up. Here's what a defensible inspection trail actually looks like.

16 May 2026

The AI WhatsApp reply that almost cost an estate agent a complaint

An AI-generated buyer reply quoted a property feature that didn't exist. The problem wasn't the model. It was the absence of a verification gate before the message left the building.

14 May 2026

The prompt library that nobody uses

Most agencies build a shared prompt library during their AI honeymoon period. Six months later it's half-broken and nobody touches it. The problem is never the library. It's the absence of ownership.

11 May 2026

Property listings can't be AI written. They can be AI assembled.

The problem with AI-generated listings isn't the AI. It's that generation without structure produces identical-sounding copy at volume. Assembly fixes that.

1 May 2026

Your CRM is a data lake. Your AI tool is drinking from the wrong tap.

Estate agents plugging AI into Reapit, Alto, or Jupix are getting confident-sounding summaries drawn from duplicate contacts, soft-deleted records, and tags nobody has touched in three years. The problem is the data, not the model.

1 May 2026

Why your agency's 24-hour turnaround promise is now a liability

Fast-turnaround guarantees made sense before AI. Now they train clients to send vague briefs faster and expect polished output anyway. Here is how to restructure the promise before it costs you.

25 Apr 2026

Why most AI pilots fail before they ship

The pattern is almost always the same: excitement, a promising first prompt, slow drift, then silent abandonment. Here are the four checks that actually break that cycle.

25 Apr 2026

Section 21 is gone: your Monday morning workflow change

Section 21 is gone. Every eviction now needs a documented evidence trail. Here's the workflow your lettings operation needs before the next possession issue.

25 Apr 2026

The brief is the bottleneck. AI can't fix bad intake

AI makes a bad brief worse, faster. This is the intake structure that gives AI tools something to work with, and cuts revision rounds in half.

25 Apr 2026

AML checks are now a workflow problem, not a paperwork problem

Having an AML policy is not the same as having an AML workflow. Here's the operational difference, and why HMRC is now testing the latter.

25 Apr 2026

Your AI reporting tool is making up your client's numbers

AI reporting tools invent numbers. Here's the verification gate your agency needs before any AI-generated client report leaves the building.

25 Apr 2026

Why AI-written property listings are a brand problem

AI-written listings all sound the same. Here's why that's a brand problem for your agency, and where AI actually belongs in your listing workflow.