17 Sept 25 
No 'ultimate list of 47 AI tools' here. Just the five that survived contact with real client work at our agency, and the three that didn't.
Every week another “ultimate list of 47 AI tools” does the rounds on LinkedIn. We’ve tried a depressing number of them. Most don’t survive two weeks of actual client work.
Here’s what did: the tools still open on our screens months later. And, more usefully, what we uninstalled.
The Five That Stayed
1. Claude, the operations layer
Started as a writing assistant; became infrastructure. It drafts briefs, audits landing pages against our CRO framework, powers chunks of our lead research, and automates a growing slice of our own agency admin. The unlock wasn’t the chat window. It was connecting it to real tools and files so output lands where work actually happens, not in a copy-paste buffer.
2. Magnific, for image generation and upscaling
Our current pick for hero images and creative exploration. The reason it beat prettier rivals: consistency. Once we locked a house style (prompt template, palette, composition rules), it reproduces that style reliably instead of reinventing itself per render. More on why that matters in our AI ad creative checklist.
3. Figma’s AI features, inside the tool we already live in
Not a separate purchase, and that’s the point. First-draft layouts, quick copy variants inside mockups, tedious renaming. Small accelerations, dozens of times a day, with zero workflow switching cost. The boring lesson: AI inside your existing tool usually beats a shinier standalone app.
4. Zoho Zia + Flow, automation with a memory
Our CRM stack runs on Zoho, for clients and for ourselves. Flow handles the plumbing (signup events to CRM to email sequences); Zia adds the intelligence layer on top: lead scoring, anomaly flags, send-time optimisation. Not glamorous. Quietly does more revenue work than any tool on this list.
5. Higgsfield, for motion and video experiments
The newest resident. Short product motion, animated variants of static winners, quick video hooks for ads. Video AI is still the wild west, and outputs need heavier human QA than image work. But ad accounts increasingly reward motion, and cutting a designer-day per video changes what’s feasible for smaller budgets.
The Three We Dropped
An AI copywriting suite (rhymes with several you’ve heard of). Fast, fluent, and relentlessly average. Every headline had the same LinkedIn-motivational cadence, and we kept catching its voice in our drafts like an accent. Writing now starts with a human angle, with AI as editor rather than author.
AI logo generators. Fine for a hackathon T-shirt. But a logo is a strategic decision compressed into a mark. The generators produce symbols with no reasoning behind them, and clients feel the hollowness within a week. Identity work stays human.
An AI website builder. Genuinely impressive demo, right up until you need the things a real business needs: performance tuning, tracking that fires correctly, CRO iterations, a CMS your team can use. We rebuild from scratch faster than we could un-bake its decisions. Our take on why speed and structure matter more than a pretty first draft covers this.
The Pattern, If You Want One
Look at the survivors: none of them replace a person. All of them remove drag around a person. A designer explores wider, a strategist audits faster, a CRM acts on signals no one had time to watch. The three we dropped all promised to replace judgement. The five we kept all amplify it.
That’s our filter, and you’re welcome to steal it: if a tool’s pitch is “no more experts needed,” run. If the pitch is “your experts, at 3x the throughput,” that one’s worth two weeks of your time.
Want the unglamorous version of this, which tool does what, wired into which client workflow? We’re happy to talk.