🌟Vasilij’s note
Anthropic CPO Mike Krieger quietly left the Figma board on 14 April. Three days later, Claude Design launched. The mainstream story has been "Figma killer." That is the wrong story for your firm. Figma was never your tool. The right story is what happens to Gamma's moat - and to your brand consistency problem - when a design system that reads your codebase once and applies it automatically to every project is bundled into a subscription you already pay for. At the same time, Salesforce tore out its own UI and made the entire platform agent-operable. OpenAI added production-grade sandboxing to its Agents SDK. And PwC published data showing that 74% of AI's economic value is being captured by just 20% of organisations - not because those firms have better tools, but because they have the discipline to move from pilots to scaled deployment. That is the advisory gap. It is also increasingly the only conversation worth having with clients.
In today's edition
This week in agents | What changed
Claude Design launches in research preview - Anthropic's first serious play at visual tooling.
Powered by Claude Opus 4.7, Anthropic's most capable vision model, Claude Design produces decks, prototypes, one-pagers, dashboards, and landing pages from prompts, codebase inputs, or uploaded design files. Available free to Pro, Max, Team, and Enterprise subscribers from 17 April. Exports to Canva, PPTX, PDF, HTML, and a handoff bundle directly to Claude Code.
Salesforce tears out its own UI - Headless 360 makes the entire platform agent-operable.
Announced 16 April at TrailblazerDX 2026, Headless 360 exposes every Salesforce capability - CRM, service, marketing, and ecommerce - as an API, MCP tool, or CLI command. Over 100 new tools and skills shipped immediately, including 60+ new MCP connectors. Agentforce Vibes 2.0 adds multi-model support, including Claude Sonnet and GPT-5.
OpenAI updates Agents SDK with native sandboxing - the governance gap narrows.
Released 15 April, the updated SDK adds isolated sandbox execution (via Cloudflare, Vercel, E2B, and Modal), a long-horizon harness for multi-step tasks, and model-agnostic support for 100+ LLMs. The Assistants API deprecation is confirmed for mid-2026.
Top moves | Signal → impact
PwC's 2026 AI Performance Study: 20% of firms capture 74% of AI's economic value - and the gap is widening.
Published 13 April, the study surveyed 1,217 senior executives across 25 sectors. The top-performing 20% of organisations are generating AI-driven financial returns 7.2 times higher than the average competitor. The differentiator is not how much AI they deploy - it is what they point it at. Leaders are using AI for growth and business reinvention. The majority are stuck in pilot mode, producing activity without measurable financial returns. → The engagement framing this creates for consultancies is direct: clients in the bottom 80% are not losing because they lack AI tools. They are losing because they lack the workflow discipline to convert pilots into scaled deployment. That is the advisory gap.Perplexity ships Personal Computer for Mac - the agent model shifts from pull to push.
Launched 16 April for Perplexity Max subscribers ($200/month), Personal Computer runs 24/7 on a user-supplied Mac mini with full access to local files, Gmail, Slack, GitHub, and native apps. It orchestrates teams of agents across 20+ frontier models and acts on triggers rather than waiting to be prompted - a structural break from every AI assistant that preceded it. Every action is auditable and reversible, with a kill switch. → The pull-to-push shift is the signal. Clients still think of AI assistants as things you ask questions to. Personal Computer is the first mainstream product that reframes the agent as something running continuously in the background, completing work between sessions. That is the mental model your clients will arrive with in six months. Prepare your advisory position now.
OpenAI commits $20 billion to Cerebras chips - inference infrastructure becomes a strategic weapon.
Reported 17 April, the deal doubles OpenAI's previously announced $10 billion Cerebras agreement and includes warrants for up to a 10% equity stake, plus $1 billion to fund Cerebras data centre construction. Cerebras simultaneously filed for IPO. The deal is explicitly about inference speed - Cerebras wafer-scale chips deliver faster, lower-latency responses than Nvidia GPUs for query workloads. → The inference war matters for consultancies building client-facing agent products. Response latency is the adoption killer for agentic workflows - users abandon agents that feel slow. OpenAI is betting $20 billion that faster inference unlocks the next wave of commercial adoption. When evaluating which model provider to build on, latency at scale is now a first-class procurement criterion alongside capability and cost.
Upskilling spotlight | Learn this week
Claude Design - Official Documentation and Get Started Guide
Covers the full capability set: design system onboarding from codebase and brand files, frontier design code-powered prototypes, export options (Canva, PPTX, PDF, HTML), and the Claude Code handoff bundle. Read this before advising clients on their visual output stack or before deploying Design in your own delivery team. The separate usage quota structure is the most important operational detail most coverage has missed.
PwC 2026 AI Performance Study - Full Report
The most important data point this week for consultancies positioning an AI offer. The 7.2x performance gap between AI leaders and laggards is the client conversation starter. The methodology - 60 AI management practices across AI use and AI foundations - gives you a diagnostic framework to structure an assessment engagement around. Read it before your next new business meeting.
Maker note | What I built this week
This week I ran Claude Design against Gamma on the same consultancy brief — a six-slide pitch deck for a fictional 38-store UK retailer, followed by a live Streamlit dashboard built from a Claude Design frontier prototype and handed off to Claude Code in a single instruction.
Decision: Claude Design does not replace Gamma for everyone. But it quietly kills Gamma's moat for consultancies that care about brand consistency and need outputs that feed into code. The dashboard demo is the moment this stops being a design tool comparison and becomes a delivery workflow comparison.
Operator’s picks | Tools to try
Claude Design
Use for brand-consistent pitch decks, client reports, interactive dashboards, and campaign one-pagers - all from a single design system configured once from your existing brand assets and codebase.
Standout: the Claude Code handoff bundle. One instruction takes a frontier design prototype to a working Streamlit app. No other deck tool on the market has an equivalent. Gamma cannot do this.
Caveat: research preview. Collaboration is basic - not fully multiplayer for 3-4 simultaneous editors. Clean brand assets required for onboarding to produce a usable design system. Separate usage quota from your standard Claude chat limits.
Gamma
Use for high-frequency marketing deck production, plug-and-play template generation, and non-technical team members who need a first draft by lunch without configuration.
Standout: speed and template variety remain best-in-class for draft-first workflows. Strong free tier. Full multi-editor collaboration today.
Caveat: no design system memory, no code handoff, no path from deck to working dashboard. $20/seat/month on Pro ($6,000/year for a 25-person team vs. £0 incremental on an existing Claude subscription).
Salesforce Headless 360 MCP Tooling
Use for building agents that read and write to client Salesforce instances without custom integration code. 60+ pre-built MCP connectors cover data, workflows, and business logic across the full platform.
Standout: Agentforce Vibes 2.0 with Claude Sonnet as default coding model, available free in the Developer Edition upgrade. Zero-friction on-ramp for consultancies building Salesforce-native agents.
Caveat: requires understanding of Salesforce's data model and permission architecture before deploying agents with write access to production orgs.
Deep dive | Claude Design vs Gamma - The Visual Output Decision for Consultancies
A typical 50-person consultancy produces 10-30 pitch decks, 5-15 client reports, 2-5 dashboards, and 20-40 social graphics every month. Each one goes through a different tool and comes back slightly off-brand. Claude Design launched on 17 April in research preview. The mainstream coverage has framed it as a Figma competitor. That is the wrong story for professional services. Figma was never your tool. The question is what Claude Design does to Gamma - and to the silent brand consistency problem that costs every firm more time and credibility than it acknowledges.
On paper
Claude Design is powered by Claude Opus 4.7, Anthropic's most capable vision model, which sees images at higher resolution and applies stronger aesthetic judgement to professional outputs. It is not an image generator. It is a visual prototyping system that builds iteratively through conversation, inline comments, and direct edits.
The design system feature is the architectural differentiator. During onboarding, Claude Design reads your codebase, design files, and existing components - and builds a system once that applies automatically to every subsequent project. Multiple systems are supported: a parent brand plus client sub-brands. This means every partner in the firm produces on-brand output from the first prompt, not after a dozen rounds of revision.
Outputs export to Canva, PPTX, PDF, HTML, and a handoff bundle directly to Claude Code - enabling a working Streamlit, React, or web application from a single design prompt. Bundled free with existing Pro, Max, Team, and Enterprise subscriptions. Separate usage quota from standard Claude chat limits.
Gamma has 70 million users, $100 million in annual revenue, and a $2.1 billion valuation. Strong free tier, $8-20/seat/month on paid plans. Default for AI-generated decks for nearly two years. Genuine wins on speed of first draft, template variety, and non-technical adoption.
In practice
On a six-slide pitch deck with the same prompt and the same brand system loaded into both tools, Gamma produces a polished output in under a minute. Template-driven. Fast. But typography drifts between slides. Colour accents do not hold across components. Stock imagery appears inconsistently. Claude Design is slower - approximately 90 seconds for first draft - but applies the design system across every element of every slide. Same font, same accent colour, same component style on slide three as on slide one.
The code handoff is where the comparison stops being close. Datadog's engineering team reduced a week-long diagnostic cycle to a single conversation using Claude Design's frontier prototyping capability. Brilliant reduced a complex interactive prototype from 20+ prompts in other tools to 2 in Claude Design. In the dashboard demo I ran for this video, a working Streamlit app with three live Plotly charts - revenue trend, store margin comparison, headcount-to-output scatter - was produced from a single Claude Design prompt and a one-instruction handoff to Claude Code. Total time: under ten minutes. Gamma has no equivalent workflow. It can style a page that looks like a dashboard. It cannot produce a working app.
For a 25-person team on Gamma Pro, the annual cost is £6,000. For the same team on an existing Claude subscription, Claude Design adds zero incremental cost.
Issues/backlash
Claude Design is research preview, not finished software. Collaboration is basic - not multiplayer for three or four simultaneous editors. Messy brand inputs produce messy outputs: firms without clean, current brand guidelines will underwhelm at onboarding and should sort that first. No mobile app. Desktop only. Features will change as the preview matures.
The design system advantage compounds only if the firm disciplines itself to onboard properly. Skipping the brand asset setup and using Claude Design as a prompt-and-generate tool produces results that are no better than Gamma - and slower on first draft.
My take (what to do)
Startup (15-40 staff): Enable Claude Design today on your existing subscription. Spend 30 minutes onboarding your brand - upload your brand guidelines PDF, your logo files, and a sample deck you consider on-brand. Run the same six-slide brief in both Claude Design and Gamma. If the design system produces consistent output on the first attempt, migrate your highest-stakes client deliverables to Claude Design and keep Gamma for internal deck churn. If the brand onboarding produces inconsistent output, fix your brand assets first - that is the real problem Claude Design is surfacing.
SMB (50-120 staff): Assign one operations team member to own the Claude Design brand system for 4 weeks. Their job: onboard the primary brand, test one client sub-brand, and document which output types (decks, one-pagers, dashboards) pass your quality bar without manual revision. Measure revision rounds per output before and after. If you are producing 15+ visual outputs per month and revision time is greater than 2 hours per output on average, Claude Design's design system will pay back within the first month. Do not roll it out firm-wide until the system is calibrated - brand drift from an incorrectly configured system is worse than the original problem.
Enterprise (150-250 staff): The dashboard workflow is your immediate priority, not the deck comparison. Map one client reporting process that currently involves manual data assembly and a static PowerPoint - your most painful quarterly close or KPI deck. Run a Claude Design → Claude Code → Streamlit pilot on that single workflow. Measure: time to produce, revision rounds, client satisfaction score. If the pilot shows a 50%+ reduction in production time, you have a productised delivery offering that no competitor currently replicates at scale.
How to try (15-minute path)
Navigate to claude.ai/design. Complete the brand onboarding with your firm's logo, brand guidelines, and one reference deck. This takes 5-7 minutes and is the only configuration you need for the design system to function. (7 min)
Paste a real brief - an actual slide prompt you would send to a junior team member - and generate a 4-6 slide output. Do not refine it. Screenshot every slide. (4 min)
Count the number of slides where typography, colour, and component style are consistent with your brand without manual revision. If more than 80% pass: this workflow is ready. If fewer than 80%: your brand asset inputs need to be cleaner - fix those before the next run. (4 min)
Success metric: a clear yes or no decision on whether Claude Design is ready for client-facing output at your firm today, based on your own brand inputs - not a vendor case study.
"Why should you ever log into Salesforce again?"
Spotlight tool | Claude Design
Bundled design system and visual prototyping tool for consultancies producing decks, reports, dashboards, and microsites. The differentiator is not generation quality - it is the design system that reads your brand once and applies it automatically to every subsequent project across every producer in the firm.
→ Design system onboarding from codebase, design files, and brand assets - applies automatically to every output
→ Frontier design: code-powered prototypes with interactive charts, voice, video, and 3D - not static mockups
→ One-instruction handoff bundle to Claude Code - from visual prototype to working Streamlit or React app
→ Full export suite: Canva, PPTX, PDF, HTML, and Claude Code handoff
→ Bundled free with existing Pro, Max, Team, and Enterprise subscriptions - separate usage quota from chat limits
What did you think of today's issue?
Did you find it useful? Or have questions? Please drop me a note., I respond to all emails. Simply reply to the newsletter or email [email protected].
This issue’s sponsor
n8n
An open‑source automation platform that lets you chain tools like DeepSeek, OpenAI, Gemini and your existing SaaS into real business workflows without paying per step. Ideal as the backbone for your first serious AI automations.
Refer and win
Share this newsletter for a chance to win!