@alex_barashkov Agreed on being more than hands. When anyone can generate a UI in minutes, what clients really pay for is judgment: what to build, what to skip, and how the process feels from their side.
@mojunel The interesting part isn't the number of agents, it's how clearly each one's job and handoffs are defined. A few well-scoped agents with a human checking the output usually beat a big swarm nobody can debug.
@rodri_shema For WhatsApp-first markets, I'd look at where invoices get ignored today rather than where payments are hardest. Brazil and Nigeria already have strong local rails, so the gap is probably the follow-up nudge, not the payment link.
@rynorhn Agreed, price is only 'too high' when the outcome is fuzzy. The moment someone can point to hours saved every week, the conversation shifts from cost to payback.
@nikhil_bansall Batching is the real lever here. One 30-minute session that turns a single idea into several post angles beats starting from a blank page every day. Consistency comes from the system, not willpower.
@hlotiim Feature lists describe the product, not the change. A launch post that says what the user can stop doing (no more manual reports, say) will out-pull "advanced automation" almost every time.
@uditsingh0909 The approve-before-it-sends step is what makes agency owners actually trust this. Drafts-first also gives you a clean place to review tone per client, which is where most AI content quietly goes wrong.
@_GeorgeReid_ Something with a done date they can point at: a launched page, a finished migration, an audit with fixes shipped. When budgets get squeezed, 'ongoing support' gets cut first. A finished thing is much harder to argue with.
@amiiinghasemi Depends on what it touches. Anything that leaves the building (a send, a refund, a client-facing reply) hits the gate every time. Internal stuff like tagging or drafting only gets flagged when a check fails or it's unsure.
The best AI model in the world just scored 12% on real work.
SemiAnalysis dropped GDPx today: 206 agentic tasks across 11 high-value fields like chip design, cybersecurity, finance and legal.
Claude Opus 5.5: 12.0%
Gemini 4 Carbon: 10.9%
GPT-6 Astra: 9.6%
On the finance and legal tasks, basically every model scored 0.
My take for agencies and freelancers:
"AI will replace you" is the wrong lesson.
The gap between 12% and "done" IS the job now. Scoping the task. Feeding the right context. Checking the output. Owning the result.
Clients won't pay for prompts.
They'll pay for whoever gets the agent from 12% to shipped.
@oliverkenyon@bysamsen Agree on the middle disappearing. The work we keep winning is the stuff where we can show how a creative changed a number, not just that we shipped ten variations.
@thesamocean Turning away 15 leads is the sign of a healthy agency. We did the same last year and our margins went up once we stopped saying yes to every project that didn't fit.
@imsonup_ One extra call vs. one email you can't unsend is an easy trade. Showing the diff instead of a yes/no prompt is the smart part, people actually read a diff.