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_]

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

[Bryan @BryanTGrime_]

https://t.co/SJsq4YTBDA

[Bryan @BryanTGrime_]

An always-on agent can only be proactive about work it understands.

Before anything runs on its own, be clear on three things. What information it can trust. What it's allowed to do. When it hands back to a person.

Skip that and it just guesses faster.

[Lee Robinson @leerob]

How do always-on, proactive agents like @Bot work?

Watch my lecture at Stanford CS146S:

2:35 How we got here
6:13 What changed in the models
10:06 Inside an always-on agent
18:07 The harness
27:26 Context engineering
34:26 Where this is going

[Bryan @BryanTGrime_]

Incredible, there isn't really more to say than that, look at the work that can get done.

[ClaudeDevs @ClaudeDevs]

We just let in every Pro and Max user from the Claude Code Projects waitlist!

If you're new to Claude Code Projects, here's a 4 minute walkthrough to get you started:

[Bryan @BryanTGrime_]

@TechGeckoHQ Without doubt the most boring problems have the most repeatable solutions that's exactly what we are solving at @Lucentframe99

[Bryan @BryanTGrime_]

@RyanWorkforceAI Perfect execution on incomplete context is still a failure.

Before an agent can act I want the policy facts in the same view as the request: what it can share, what it cannot, and when to stop. Otherwise it does what you asked and still creates a mess.

[Bryan @BryanTGrime_]

@cloudwithraj Treating the model as a separate layer is how demos look clever and live work breaks.

The useful question is which operational systems it can trust, and which decisions still need a human before anything goes to a customer. Wire those first. Then decide where a model belongs.

[Bryan @BryanTGrime_]

Faster feature shipping is useful until the firm cannot absorb the change.

In a trades business the bottleneck is rarely the SaaS release calendar. It is whether enquiry, photos, quote approval and callback still live in one place. New tools on top of a split inbox just move the same leak.

[Bryan @BryanTGrime_]

@andywalkerhq The third group wins because follow-up is a system, not a mood.

More leads into a broken callback process just grows the ignored pile. I would fix same-day contact and a clear next action on every enquiry before spending another pound on acquisition.

[Bryan @BryanTGrime_]

@FleetOpsCo Temporary with no TTL is how permissions go permanent. No owner is how nobody notices.

Same rule I use for sales ops tools: every widen needs an expiry and a named human who can revoke it. Debug debt turns into production risk quietly.

[Bryan @BryanTGrime_]

@tomedwards Writing the list first is the useful step. Frequency ranking beats vibes.

I do the same inside my own operating business: name the recurring admin, then decide what a person still owns. Delegation without that list just moves the chaos onto someone else.

[Bryan @BryanTGrime_]

The ledger is territory. The agent that posts into it is the operating system.

Same pattern I see in service firms: whoever owns the workflow that moves work between systems owns the customer relationship. The app brand matters less than who decides what the agent is allowed to do.

[Bryan @BryanTGrime_]

@LimestoneHQ That list is the order most teams reverse. Model first is how you automate confusion.

I start with one painful workflow, name who can approve and escalate, then map the workarounds. Without those three, cycle-time baselines just measure a mess more accurately.

[Bryan @BryanTGrime_]

That question is the real gate. If the only signal is a customer calling angry, you find out too late.

I want an unsure queue a non-technical person can work, and a short note on what "wrong" looks like, before the agent ever touches a live lead. Boring, and it is what keeps the firm owning the system.

[Bryan @BryanTGrime_]

That's exactly the failure mode. Website form in one place, WhatsApp photos in another, and the person picking up the phone has half the job.

Inside my own operating business I'm pulling both into one enquiry view so the caller sees the form details and the photos together before they prepare a quote. Same enquiry, same next action. Still unfinished, but that join is the bit that matters.

[Bryan @BryanTGrime_]

Most businesses don't need a clever agent first. They need to see the work.

Inside my own operating business, Revitalize Resprays, website enquiries landed in one place and WhatsApp enquiries landed in another. When you're busy on jobs, that split costs you. You don't know who to call first, and good leads go cold.

So before automating anything, I pulled both into one lead view. Each enquiry shows where it came from, the service, the timeline, a lead grade and the next action.

It isn't finished. Next I'm tightening the handoff from new lead to qualification to call priority to an approved quote.

The rule stays the same. Capture the right information. Make the next action obvious. Keep a human on pricing, customer contact and commitments.

Boring problem. Commercially important. That's where the time comes back.

[Bryan @BryanTGrime_]

I'm having multiple conversations with business owners daily, who all struggle with the boring, repetitive tasks that become deceivingly long and momentous, this is what I look forward to solving most, forward deploying agents to conquer time back for businesses, Human led.

[Bryan @BryanTGrime_]

@Subhorovroy Tools on top of conflicting lists just move the argument. One source of truth first, then automation. Same pattern in trades: one lead record beats three inboxes arguing about who called back.

[Bryan @BryanTGrime_]

@anmoltt That order is useful. Owners often blame the ads when the leak is after the form. I look at same-day contact rate before I touch creative. If intake is slow, more leads just grow the ignored pile.

[Bryan @BryanTGrime_]

@Hussein_mahrz Capture and route is the first half. The second half is a "not ready yet" branch: missing postcode, missing service, missing timeline. Park those, ask one question, keep a human on the go/no-go before anyone quotes.