Bryan

Founder of Lucent Frame. I audit, build and deploy practical AI systems inside UK businesses. Sharing what works and what breaks in real operations. Manchester.

[Bryan @BryanTGrime_]

I'm Bryan. I'm a Manchester-based operator and I'm building Lucent Frame.

Lucent Frame helps businesses understand where AI can genuinely improve the way they work. Then we design and build practical systems around those problems.

We don't start with a chatbot. We don't start with software. We start with the business.

I call that Phase 0. Before anything gets built, you need to know how work comes in, who handles it, where the information lives and where it gets stuck. If you don't understand the process, you can't safely automate it.

Why I'm doing this. I've spent years in sales, cold calling, lead generation and running local service businesses. I know what it's like when leads go cold, quotes sit unsent and customer details live in five different places. A lot of AI advice ignores that reality.

I also run Revitalize Resprays, a kitchen and bedroom respraying business. It's my own operating business, not a client. I'm using it as a live proof environment for lead capture, follow-up, quoting and CRM workflows. Real process, not a demo.

If you follow along, you'll get:

Plain explanations of AI for people who don't work in tech
What I'm learning building systems inside a real trades business
What works, what breaks and what I'd do differently
Where the human still needs to make the call

No hype. No made-up results. Just the honest version.

Human led. Be the human.

[Bryan @BryanTGrime_]

@_shaunirvine I think the only way we get out of this is to go down on camera, like when I say that, I mean it documented inside and out, we need to go down on this documentary, otherwise the club will never be free from this type of owner, we're literally just a buy to let.

[Bryan @BryanTGrime_]

@violetainsights Yes. If nobody can say what was kept, why, and when it expires, you do not have memory. You have an unowned pile.

Write retention and a named owner before you let an agent store anything from customer work.

[Bryan @BryanTGrime_]

@NOVYQHoldings Extraction without a decision gate just speeds up filing.

Keep the source evidence, a human approval on pricing or supplier choice, and an audit trail. That is when it becomes decision support rather than busywork.

[Bryan @BryanTGrime_]

@usecompassapp That sign flip is the warning. Confidence in the tool without confidence in your own check becomes rubber-stamping.

I keep AI on the repetitive draft. Price, commit, and anything a customer will treat as a promise stay with a person.

[Bryan @BryanTGrime_]

@chams_builds That speed is real. A small firm can try one change on Monday and know by Friday.

I would still write what the system must never do alone before you switch it on. Fast experiments still need a stop path.

[Bryan @BryanTGrime_]

@AppollonGAI That is the commercial point. Owners buy fewer missed enquiries and a fuller week, not another AI label.

Sell the outcome and the process behind it. The tool is how you get there, not the pitch.

[Bryan @BryanTGrime_]

@mia_ai_a That order is right. Strip secrets and personal data before anything hits a tool.

In a small firm the risk is often a paste into the wrong chat, not a fancy breach. Data minimization is the boring gate that actually gets used.

[Bryan @BryanTGrime_]

@DavidAgentic Controls outside the model is the bit most demos skip.

Human approval on anything that can change the live system, a sandbox for experiments, and a log someone can read without the builder. That is what makes an agent ownable by the firm.

[Bryan @BryanTGrime_]

@deeponailabs Showing the plan before it acts is the trust move.

I want the same on business agents: proposed next step visible, human gate on anything that commits money or a customer, then run. Act first and explain later is how you lose the room.

[Bryan @BryanTGrime_]

Before you switch an agent on, name one person who owns it when it goes wrong.

And give them a stop button they can use without calling IT.

If nobody owns the failure path, you do not have a system. You have a demo.

[Bryan @BryanTGrime_]

@taointelligence Silent field blanks are the worst kind of bug.

I treat schema and permissions as separate checks now. The API accepting a shape is not the same as the role being allowed to write it.

[Bryan @BryanTGrime_]

@Matyas_Jirsa Understanding the bottleneck is the right start.

I would still write what the system must never do on its own before you automate it. The boring gate is what keeps a quick win from becoming a messy cleanup.

[Bryan @BryanTGrime_]

@Amagi1234567 Consent fatigue is real in small firms too.

If every low-risk step needs a click, people stop reading. Keep human approval for the high-impact stuff: price, customer messages that commit you, and anything hard to undo.

[Bryan @BryanTGrime_]

@deanwithani Agree. Autopilot tools without a human gate just speed up the wrong work.

I use AI for the repetitive bits. Price, commit, and anything a customer will treat as a promise stay with me.

[Bryan @BryanTGrime_]

@bbelderbos Most of the reliability work sits outside the model.

Write the inputs it can trust, the checks before anything goes live, and who owns a bad answer. The model fills the messy middle. The system around it keeps the business safe.

[Bryan @BryanTGrime_]

@TechGeckoHQ @Lucentframe99 Lucent Frame is the firm I am building around that.

We audit how work actually moves in UK businesses, then design and deploy practical systems so repetitive commercially important tasks stop eating the day. Human led. Clarity before technology.

[Bryan @BryanTGrime_]

@buildznofficial Buying the tool without a clear one-pager of how work moves just moves the same fuzzy handoffs into new software.

Process first. Tool second.

[Bryan @BryanTGrime_]

Most firms buy a tool, then try to force the work into it.

Flip it.

Write what "ready to quote" means on one page: required fields, who calls, who can approve a price, what happens if info is missing.

If that page is fuzzy, any model or workflow will invent the gaps.

Process first. Then the tools.

[Bryan @BryanTGrime_]

@RyanWorkforceAI That is the gap. If agent managers cannot see what was allowed, what was shared, and when to stop, governance stays a slide deck.

The useful view shows the request, the policy beside it, and a stop path a non-technical person can use.

[Bryan @BryanTGrime_]

Before an agent writes to your CRM, name the fields it must never touch.

Price. Anything that commits the job. Anything a customer could treat as a promise.

The agent can draft. A person owns the live record.

[Bryan @BryanTGrime_]

@TTrimoreau They won’t they’ll hire a FDE