Claude Opus 4.8 shipped this week with Dynamic Workflows, a research preview in Claude Code that changes how multi-step work gets done. Instead of you prompting back and forth to complete a complex task, Claude now plans the work itself, spins up parallel sub-agents to tackle different parts simultaneously, verifies the outputs, and delivers a finished result.

For marketers and founders who already use Claude Code, this is one of the most practically useful updates in months. Here is how to put it to work immediately.

What Dynamic Workflows Actually Does

In standard Claude Code usage, you prompt, Claude responds, you review, you prompt again. Each step waits for you.

Dynamic Workflows changes this. You hand Claude a complex goal. Claude breaks it into sub-tasks, assigns each to a parallel sub-agent - one agent researches, another outlines, another drafts, another fact-checks - then reassembles the outputs into a single verified result. You set it running and come back to a finished product.

The practical difference: instead of briefing one freelancer one task at a time, you brief a small coordinated team on the full objective in one go.

How to Access Dynamic Workflows

Dynamic Workflows is currently in research preview inside Claude Code. To access it:

  1. Install or update Claude Code: npm install -g @anthropic-ai/claude-code
  2. Open a project directory and run claude
  3. Describe a multi-step task - Claude will plan and run it with Dynamic Workflows when the task is complex enough to benefit from parallelization

You need an active Claude API key or a Claude Max subscription with Claude Code access.

Task 1: Competitive Content Audit

The prompt: "Run a competitive content audit for [your niche]. Research the top 5 competitors in [market], identify the top 10 topics each covers, compare their coverage against my current content (attached), flag the 5 gaps with the highest traffic potential, and produce a prioritized content calendar for the next 8 weeks."

Why it works here: This task has 5 parallel research tracks (one per competitor), a comparison layer, and a synthesis step. Without parallel sub-agents, you'd run each competitor separately across multiple sessions. Dynamic Workflows runs all 5 simultaneously and delivers one merged output.

Task 2: Campaign Ideation and Briefing

The prompt: "Generate a campaign for [product/offer]. Create 3 distinct concepts each targeting a different audience segment, write a full brief for the strongest concept including: audience definition, messaging framework, 5 ad headlines, 3 email subject line options, and a 30-day content calendar. Evaluate each concept against [your stated goal] before selecting the strongest."

Why it works here: The 3 concept tracks run in parallel. An evaluation agent reviews all 3 before Claude selects the winner - so you receive the strongest one with a rationale already built in, rather than reviewing three concepts yourself.

Task 3: SEO Content Briefs at Scale

The prompt: "Create SEO content briefs for the following 5 keywords: [list]. For each, research top-ranking competitors, identify the intent, extract the key subtopics covered, draft a full H1/H2/H3 outline, suggest 3 internal linking opportunities, and write a 150-word meta description."

Why it works here: Each keyword gets its own sub-agent. Five briefs run in parallel rather than sequentially. What normally takes 2-3 hours of back-and-forth becomes a single session you step away from and return to.

Task 4: Weekly Performance Report

The prompt: "Write this week's marketing performance report. Pull the data from [attached CSV exports from Google Ads, Google Analytics, and your email platform]. Identify the 3 biggest wins, 3 biggest underperformers, the root cause of each, and recommended actions for next week. Format as an executive summary ready to share with stakeholders."

Why it works here: Different sub-agents analyze different data sources at the same time - one on ads, one on web analytics, one on email - before a synthesis agent assembles the narrative. The report you receive is already integrated, not three separate channel outputs you have to combine manually.

Task 5: Voice-of-Customer Research Synthesis

The prompt: "Synthesize audience research from the following inputs: [attach customer interview transcripts, G2/Capterra reviews CSV, and NPS survey data]. Identify the top 5 customer pain points, the language customers use to describe each problem, the most common objections to [product], and the 3 most frequently mentioned desired outcomes. Output a 2-page voice-of-customer summary I can use for copywriting."

Why it works here: Each data source runs through its own sub-agent simultaneously. The synthesis agent merges the themes and resolves contradictions across sources. You get one integrated document instead of three separate summaries to reconcile yourself.

The Bottom Line

Dynamic Workflows does not replace your judgment - it handles the parallelization and coordination that slow your work down. The five tasks above represent the kinds of work where you spend the most time managing the process rather than making decisions. That is exactly what parallel sub-agents are built for.

The research preview is live now in Claude Code. If you already run Claude Code in your workflows, this is worth testing this week.