Perplexity's Personal Computer and Custom Skills Era: How UK Marketing Teams Should Respond

100M+ — Perplexity monthly active users as of May 2026  ·  26 May 2026 — xAI Custom Skills for Grok official launch  ·  1M tokens — Grok 4.3 context window with native video input  ·  22-23 May — Perplexity Personal Computer week of launch with Mac mini hardware

There are weeks where the AI roadmap shifts by a feature, and there are weeks where it shifts by a category. The last fortnight of May 2026 belonged to the second kind. Perplexity launched Personal Computer — an always-on AI running on a dedicated Mac mini, configured as a 24/7 digital proxy of you, with explicit approval flows, a full audit trail and a physical kill switch — alongside Workflows for market research, sales prep, slide creation and SEO site audits, Space Skills for shareable team capabilities, and live-warehouse connectors into Snowflake and Databricks. The company simultaneously crossed 100 million monthly active users. Three days later, on 26 May 2026, xAI shipped Custom Skills for Grok officially, sitting on top of Grok 4.3 with its 1M-token context and native video input, the Grok Build 0.1 coding model public beta on 29 May, and Custom Voices for voice cloning.

Read individually, these are interesting product announcements. Read together, they describe the end of prompt-and-pray and the start of always-on agentic Workflow automation as the default way marketing teams ship work. The model is no longer open a chat tab, type a question, paste an output. The model is build a Skill or a Workflow once, let it run on a cadence, review what it produces against an audit log, and intervene only on exceptions. For a Head of Marketing or a CMO running a UK team — whether D2C, B2B SaaS, professional services or retail — that shift changes how many initiatives you can credibly run in parallel, how lean your team can be, and how much of your day is spent reviewing rather than producing.

Per our standard editorial cough, we will declare an interest. BraivIQ designs and deploys exactly this kind of Agentic AI London infrastructure for mid-market UK clients, so we benefit when more marketing leaders take this seriously. We have tried to write the piece we would want to read if we were on the buying side rather than the building side: specific about what the tools do, specific about what they do not do, and specific about how to avoid getting trapped in someone else's roadmap.

Why the Shift from Prompt-and-Pray to Always-on Workflows Matters

For three years, the standard interaction model with generative AI was synchronous. You opened a tab, you typed a question, you got an answer, and you copied bits of it into a deck. That model never really matched how marketing teams work. A marketing team's value is not in producing one piece of analysis on demand — it is in producing dozens of recurring pieces of analysis, content, intel and decisions every week, against a calendar, with quality control. Synchronous chat does not fit that shape.

Perplexity Personal Computer and Grok Custom Skills both ship the same insight from different angles. Personal Computer makes the agent always available, on its own hardware, watching its own queue, executing approved tasks across your tools, with the audit log running the whole time. Custom Skills makes the high-leverage prompts reusable — you build the how we write our quarterly investor update Skill once, and any team member runs it daily without reinventing the prompt. Both moves push the burden away from individual prompting talent and onto a system that can be inspected, governed and improved. The marketing leader's job changes accordingly: less time spent generating outputs, more time spent designing Workflows, reviewing exceptions, and owning the brand-safety perimeter. AI Automation, deployed well, finally separates busyness from output.

What Perplexity Workflows and Grok Custom Skills Actually Replace in a Marketing Week

Weekly competitive teardowns and SEO content audits

Most marketing teams do competitive teardowns badly because they are laborious. A Perplexity Workflow scheduled every Monday morning can crawl your top six named competitors' homepages, pricing pages, blog posts and LinkedIn output, summarise changes since last week, flag pricing or positioning shifts, and drop a structured brief into your shared Space before stand-up. With live Snowflake and Databricks connectors, you can join that intel to your own pipeline data so the brief tells you which deals or campaigns the change actually affects. The Workflow for SEO site audit replaces a quarterly agency engagement with a weekly automated pass: which pages have decayed, where competitor pages have leapfrogged you, what the topical clusters look like now, and what the next ten briefs should be. Combine this with Grok 4.3's 1M-token context and a Custom Skill drafts the rewrites in your house voice.

ABM research, social listening and reply triage

Sales prep Workflows can run nightly across your named account list, ingesting filings, news, hires, product launches and exec moves, surfacing the five accounts most likely to be in-market this week. For a B2B SaaS team running ABM, this collapses a research function that historically took two FTEs into a review function that takes one Head of Demand half a day. On the social side, Grok's native integration with X plus Custom Skills gives you a triage layer: a Skill that monitors brand mentions, classifies by severity, drafts reply candidates in your tone of voice, and queues them for a human approval. You are not handing the keys to a model — you are eliminating the 80 percent of triage volume that is repetitive.

Slide and deck generation

The Workflows for slide creation are the most obviously useful for execs. A monthly board update Workflow that pulls from your warehouse, your CRM, your campaign analytics and your brand template, and produces a v1 deck that a CMO edits rather than builds — that is two days a month of CMO time recovered, every month. Multiply that across quarterly business reviews, pitch decks and investor updates and the recovered hours pay for the entire infrastructure investment in the first quarter.

A Realistic Case Sketch: London D2C Brand

Picture a London-based direct-to-consumer brand doing around 40 million in annual revenue across skincare and wellness. Marketing team of eleven. Two recurring problems: competitor pricing moves faster than they can react, and customer-service-adjacent social replies are eating a junior community manager's full week. Workflow one is a Perplexity Workflow scheduled to run every weekday at 06:30. It pulls competitor pricing across twelve named SKUs from twenty-three retailer sites, joins it to their own pricing data in Snowflake, flags any move greater than five percent, and drafts a tactical response brief (price match, promo, hold) for the brand director's 09:00 stand-up. The audit trail records every source page and every reasoning step. The kill switch is a physical button on the Mac mini in the office.

Workflow two is a Grok Custom Skill called Reply Triage v3. It watches mentions across X and Instagram, classifies them into five buckets, drafts a reply in the brand's documented voice (built once, version-controlled), and queues drafts in a dashboard the community manager clears in forty minutes a day instead of five hours. Hard refusals are routed to customer service via a Slack channel, with no auto-send. The community manager now spends the recovered time on original content and creator partnerships — actual marketing work. Neither Workflow makes the brand AI-native. Both Workflows shift a meaningful number of FTE-hours per week from production to judgement. That is the realistic shape of Agentic AI London deployments in 2026 — not the science-fiction version, but a measurable shift in where the team's time goes.

Brand Safety, Hallucination and Audit Trails: What to Gate Before You Ship

The temptation with always-on agentic tooling is to wire it up to publishing channels because the tooling allows it. Resist that. The single biggest reputational risk for a UK consumer or B2B brand in 2026 is not slow adoption of AI — it is an unattended agent posting something wrong, off-brand or non-compliant under the brand handle. Three gates are non-negotiable. First, an explicit approval surface for anything customer-facing — no auto-send to public channels, ever, without a human click. Second, an audit log that records the prompt, the model output, the sources cited, the approval action and the human approver, retained for at least twenty-four months for ICO and regulatory readiness. Third, a hallucination guardrail — for factual claims (pricing, regulation, medical, financial), require the Workflow to cite the underlying source and refuse if the source is not present or is older than a defined threshold.

On the brand voice side, build a Custom Skill or Space Skill that encodes your editorial standards — words you use, words you do not, claims you can make, claims you cannot — and route every public-facing draft through it as a final pass. Treat this as the editorial equivalent of a pre-flight checklist. Pair this with a quarterly red-team exercise where someone deliberately tries to make the agent produce something off-brand or non-compliant. The output of that exercise becomes the next iteration of the guardrails.

Org-Design: Who Owns the Workflow When It Crosses Departments

Most of these Workflows do not sit cleanly inside marketing. Competitor pricing Workflows touch revenue ops. ABM research Workflows touch sales. Reply triage Workflows touch customer success. Slide generation Workflows touch finance and the exec team. If you stand up Workflow automation without a clear owner, you get the standard cross-functional outcome: nobody maintains it, drift sets in, the outputs degrade, and within six months the team stops trusting them. The pattern that works is a named Head of AI Operations or a dotted-line role inside marketing ops with explicit authority to commission, maintain and retire Workflows across the marketing-adjacent surface. Where the role does not yet exist, a fractional engagement with an AI Agency London partner that runs the same function gives you the muscle without the headcount. Either way, ownership has to be explicit. Shared ownership of a Workflow is no ownership of a Workflow.

The 90-Day UK Marketing Team Playbook

1. Days 1-15: Audit your recurring outputs. List every artefact your marketing team produces on a weekly or fortnightly cadence. Score each by hours consumed and judgement required. Anything high-hours and low-judgement is a Workflow candidate. Pick three for the pilot.
2. Days 16-30: Stand up the infrastructure. Choose at least two vendors (a typical pairing is Perplexity for research-heavy Workflows and Grok or Claude for content-heavy Custom Skills) to avoid single-vendor lock-in. Provision the Mac mini or equivalent always-on surface. Wire warehouse connectors. Define your audit log retention policy and your approval surface.
3. Days 31-60: Build the three pilot Workflows in production. Run them in parallel with the existing manual process for thirty days. Measure: time saved, error rate, brand-safety incidents (target zero), reviewer satisfaction. Iterate weekly with named owners.
4. Days 61-75: Retire the manual versions. If the Workflows are passing review, stop doing the work manually. This is the step teams skip — they end up doing the work twice and burning out. Commit to the retirement.
5. Days 76-90: Productise into Space Skills and Custom Skills. Package the patterns into shareable Skills so other team members and other departments can deploy similar Workflows without reinventing them. Brief the exec team on results, fund the next wave of three Workflows, and name the permanent owner role.

Sources

1. Perplexity Changelog, Introducing Personal Computer, Workflows and Space Skills, May 2026
2. VentureBeat, Perplexity launches always-on AI Personal Computer with audit trail and kill switch, May 2026
3. basenor.com, xAI weekly roundup: Custom Skills for Grok, Grok 4.3 and Build 0.1 beta, 27 May 2026
4. Reuters, Perplexity crosses 100 million monthly active users as AI search competition intensifies, May 2026
5. Bloomberg, xAI ships Custom Skills as Musk pushes Grok into enterprise workflows, 26 May 2026
6. The Financial Times, Always-on AI agents reach UK marketing teams as Perplexity and xAI race for the desktop, May 2026
7. TechCrunch, Grok 4.3 lands with 1M-token context and native video input, May 2026
8. IT Pro, What Perplexity Workflows mean for UK enterprise buyers, May 2026
9. Computer Weekly, Agentic AI moves from chat to scheduled execution: what CIOs should ask vendors, May 2026
10. MIT Technology Review, The shift from prompt-and-pray to always-on agents, May 2026
11. The Guardian, AI agents in the marketing department: productivity boost or brand-safety risk, May 2026
12. Information Commissioner's Office (ICO), Guidance on audit trails and human oversight for generative AI in customer-facing contexts, 2026
13. Competition and Markets Authority (CMA), AI Foundation Models: update on competition and consumer protection priorities, 2026