Agent Use Cases

Autonomous agents represent a fundamental shift in how software interacts with the web. Unlike humans who make 5-10 API calls per hour, agents can orchestrate hundreds or thousands of requests per session. The agentic browser with x402 support enables this high-frequency, autonomous operation while maintaining user control over spending and security.

Why Agents Need Specialized Browsers

Traditional browsers were designed for human interaction: manual authentication, visual rendering, cookie-based sessions. Autonomous agents have different requirements:

  • Cryptographic identity: Agents use threshold signatures, not passwords
  • Programmatic access: HTTP-native payments (x402), not credit card forms
  • High-frequency transactions: 100-1000 API calls/hour, not 5-10
  • Spending controls: Cryptographic limits ($100/day), not software guardrails
  • Self-custody: User owns keys, agent holds 1-of-3 share

The following use cases demonstrate how the agentic browser transforms agent capabilities across research, trading, and content creation workflows.

Agent Persona 1: Research Agent

Workflow Overview

A research agent autonomously searches, synthesizes, and fact-checks information across multiple sources. Typical tasks include academic literature review, market research, competitive analysis, and data aggregation.

┌──────────────────────────────────────────────────────────┐
│              RESEARCH AGENT SESSION                      │
├──────────────────────────────────────────────────────────┤
│                                                          │
│  1. User Query: "Analyze threshold signature research   │
│                  published in 2024-2025"                 │
│                                                          │
│  2. Agent Workflow:                                      │
│                                                          │
│     ┌─────────────────┐                                 │
│     │ Search APIs     │  ←  10 searches across academic │
│     │                 │     databases (arXiv, IEEE,     │
│     │ $0.02 × 10      │     Google Scholar)             │
│     └────────┬────────┘     Total: $0.20                │
│              │                                           │
│              v                                           │
│     ┌─────────────────┐                                 │
│     │ Paper Access    │  ←  15 PDFs downloaded          │
│     │                 │     ($0.50 each via x402)       │
│     │ $0.50 × 15      │     Total: $7.50                │
│     └────────┬────────┘                                 │
│              │                                           │
│              v                                           │
│     ┌─────────────────┐                                 │
│     │ LLM Synthesis   │  ←  200k tokens processed       │
│     │                 │     (GPT-4: $2/1M in, $10/1M out)│
│     │ $2.40 total     │     Total: $2.40                │
│     └────────┬────────┘                                 │
│              │                                           │
│              v                                           │
│     ┌─────────────────┐                                 │
│     │ Fact-Check APIs │  ←  5 verification calls        │
│     │                 │     ($0.10 each)                │
│     │ $0.10 × 5       │     Total: $0.50                │
│     └────────┬────────┘                                 │
│              │                                           │
│              v                                           │
│  3. Total Session Cost: $10.60                          │
│     Time: 3 minutes (vs. 4 hours manual)                │
│                                                          │
│  4. Agent Authorization:                                │
│     • Threshold: $20/session auto-approved              │
│     • User notification: Summary sent, no action needed │
│                                                          │
└──────────────────────────────────────────────────────────┘

Research Agent Workflow

Cost Breakdown

ServiceCallsCost/CallTotalNotes
Academic Search APIs10$0.02$0.20arXiv, IEEE, Google Scholar
PDF Access (x402 paywalls)15$0.50$7.50Per-article micropayment
GPT-4 Synthesis200k tokens$0.012/1k$2.40100k in ($2), 100k out ($10/1M)
Fact-Check APIs5$0.10$0.50Citation verification
Gas Fees (Solana)30 txs$0.0001$0.003With payment channel: 2 TX only

Total: $10.60 for a comprehensive research report that would take a human 4+ hours to compile. The agent completes it in 3 minutes.

ROI Analysis

Assume a researcher's time is valued at $50/hour. Manual research takes 4 hours ($200 in labor). Agent-assisted research:

  • Agent cost: $10.60
  • Human oversight: 15 minutes ($12.50)
  • Total cost: $23.10
  • Savings: $176.90 per report (88% reduction)

At 10 reports per month, annual savings: $21,228.

Code Example

research-agent.ts
import { AgenticBrowser } from '@agentic-browser/sdk';

class ResearchAgent {
  constructor(
    private browser: AgenticBrowser,
    private spendingPolicy: SpendingPolicy
  ) {
    // Configure auto-approval threshold
    this.spendingPolicy.setSessionLimit(20_00); // $20 per session
    this.spendingPolicy.setDailyLimit(100_00);  // $100 per day
  }

  async researchTopic(query: string): Promise<ResearchReport> {
    console.log(`Starting research: "${query}"`);

    // Step 1: Search academic databases (x402 APIs)
    const searchResults = await Promise.all([
      this.searchArXiv(query),
      this.searchIEEE(query),
      this.searchGoogleScholar(query),
    ]);

    // Automatically approved: $0.02 × 10 = $0.20 < session limit
    const papers = searchResults.flat().slice(0, 15);

    // Step 2: Access full-text papers (x402 paywalls)
    const fullTexts = await Promise.all(
      papers.map(paper => this.accessPaper(paper.doi))
    );
    // Cost: $0.50 × 15 = $7.50 (auto-approved, under session limit)

    // Step 3: LLM synthesis
    const synthesis = await this.synthesizeFindings(fullTexts);
    // Cost: ~$2.40 for 200k tokens (GPT-4)

    // Step 4: Fact-check claims
    const factChecked = await this.verifyClai ms(synthesis.claims);
    // Cost: $0.10 × 5 = $0.50

    // Total: $10.60 (within $20 session limit, no user prompt)
    return {
      query,
      papers,
      synthesis,
      factChecked,
      totalCost: 10.60,
      timestamp: Date.now(),
    };
  }

  private async searchArXiv(query: string): Promise<Paper[]> {
    // x402 payment handled automatically by browser
    const response = await this.browser.fetch('https://api.arxiv.org/search', {
      method: 'POST',
      body: JSON.stringify({ query, max_results: 5 }),
      // Browser intercepts 402 response, signs payment with threshold wallet
    });

    return response.json();
  }

  private async accessPaper(doi: string): Promise<string> {
    // Publisher returns 402 Payment Required
    // Browser pays $0.50 via x402, receives full-text PDF
    const response = await this.browser.fetch(`https://doi.org/${doi}`);
    return response.text();
  }

  private async synthesizeFindings(texts: string[]): Promise<Synthesis> {
    const combined = texts.join('\n\n');

    // OpenRouter x402 API (cost-basis pricing via >< token)
    const response = await this.browser.fetch('https://openrouter.ai/api/v1/chat', {
      method: 'POST',
      headers: { 'X-Payment-Token': '><' },
      body: JSON.stringify({
        model: 'openai/gpt-4-turbo',
        messages: [{
          role: 'user',
          content: `Synthesize key findings from these papers: ${combined.slice(0, 50000)}`
        }],
        max_tokens: 2000,
      }),
    });

    return response.json();
  }

  private async verifyClaims(claims: string[]): Promise<FactCheck[]> {
    // Fact-check API with x402 micropayment
    return Promise.all(
      claims.map(claim =>
        this.browser.fetch('https://factcheck.ai/verify', {
          method: 'POST',
          body: JSON.stringify({ claim }),
        }).then(r => r.json())
      )
    );
  }
}

// Usage
const agent = new ResearchAgent(browser, spendingPolicy);

const report = await agent.researchTopic(
  'Threshold signatures for MPC wallets: GG20 vs CGGMP21 comparison'
);

console.log(`Research complete. Cost: $${report.totalCost}`);
console.log(`Found ${report.papers.length} papers`);
console.log(`Synthesis: ${report.synthesis.summary}`);

Agent Persona 2: Trading Agent

Workflow Overview

A trading agent monitors market data, executes algorithmic strategies, and manages portfolio risk. Unlike research agents, trading agents require sub-second latency and high transaction throughput.

┌──────────────────────────────────────────────────────────┐
│           TRADING AGENT: 10-MINUTE SESSION               │
├──────────────────────────────────────────────────────────┤
│                                                          │
│  Strategy: Market-making on SOL/USDC pair               │
│                                                          │
│  1. Market Data Streaming:                               │
│     ┌──────────────┐                                     │
│     │ Price Feed   │  ←  600 updates (1/second)         │
│     │ API          │     $0.0001 per update             │
│     └──────────────┘     Total: $0.06                   │
│                                                          │
│  2. Order Book Snapshots:                                │
│     ┌──────────────┐                                     │
│     │ Depth Data   │  ←  60 requests (1 per 10s)        │
│     │ API          │     $0.001 per request             │
│     └──────────────┘     Total: $0.06                   │
│                                                          │
│  3. Trade Executions:                                    │
│     ┌──────────────┐                                     │
│     │ DEX API      │  ←  25 trades (market making)      │
│     │ (Jupiter)    │     $0.01 per trade                │
│     └──────────────┘     Total: $0.25                   │
│                                                          │
│  4. Risk Analysis:                                       │
│     ┌──────────────┐                                     │
│     │ Portfolio    │  ←  10 risk calculations           │
│     │ Analytics    │     $0.02 per calculation          │
│     └──────────────┘     Total: $0.20                   │
│                                                          │
│  Total API Cost: $0.57 for 695 API calls                │
│  Gas (with payment channel): $0.0002 (2 TX)             │
│                                                          │
│  Trading PnL: +$15.40 (spread capture)                  │
│  Net Profit: $14.83 per 10 minutes                      │
│               = $89/hour = $712/day (8 hours)           │
│                                                          │
└──────────────────────────────────────────────────────────┘

Trading Agent Workflow (10 minutes)

Cost Breakdown

ServiceFrequency10-Min SessionCost/CallTotal
Real-time Price Feed1/second600 updates$0.0001$0.06
Order Book Depth1/10 seconds60 snapshots$0.001$0.06
Trade Execution (DEX API)~2.5/minute25 trades$0.01$0.25
Risk Analytics1/minute10 calculations$0.02$0.20
Gas (Channel)Open + Close2 TX$0.0001$0.0002

Total Session Cost: $0.5702

Without payment channels, 695 on-chain transactions would cost $0.0695 in gas—a 12% overhead. Payment channels reduce this to 0.035% overhead.

ROI Analysis

  • API costs: $0.57 per 10 minutes
  • Trading PnL: +$15.40 (spread capture from 25 market-making trades)
  • Net profit: $14.83 per 10 minutes
  • Hourly rate: $89/hour
  • Daily (8 hours): $712
  • Monthly (22 days): $15,664

The agent's API costs are 3.7% of revenue. This is only economically viable because:

  1. x402 enables sub-cent micropayments (no $0.30 credit card fees)
  2. Payment channels eliminate per-transaction gas overhead
  3. Threshold wallet allows agent to auto-sign within user-set limits

Code Example

trading-agent.ts
import { AgenticBrowser } from '@agentic-browser/sdk';
import { PaymentChannelClient } from '@x402/solana';

class TradingAgent {
  private channel: PaymentChannelClient;

  constructor(
    private browser: AgenticBrowser,
    private spendingPolicy: SpendingPolicy
  ) {
    // High-frequency trading needs payment channels
    this.spendingPolicy.setHourlyLimit(10_00);  // $10/hour
    this.spendingPolicy.setPerTxLimit(0_10);    // $0.10 per transaction
  }

  async runMarketMakingSession(durationMs: number): Promise<SessionReport> {
    // Open payment channel with Jupiter API (off-chain updates)
    this.channel = await this.browser.payments.openChannel(
      JUPITER_API_PROVIDER,
      100_000_000n  // 100 USDC initial deposit
    );

    const startTime = Date.now();
    const trades: Trade[] = [];
    let apiCalls = 0;

    // Main trading loop
    while (Date.now() - startTime < durationMs) {
      // 1. Stream price data (x402 WebSocket, paid per message)
      const price = await this.getPriceUpdate();  // $0.0001
      apiCalls++;

      // 2. Fetch order book depth every 10 seconds
      if (apiCalls % 10 === 0) {
        const depth = await this.getOrderBook();  // $0.001
        apiCalls++;
      }

      // 3. Execute market-making strategy
      const signal = this.calculateSignal(price, depth);
      if (signal.shouldTrade) {
        const trade = await this.executeTrade(signal);  // $0.01
        trades.push(trade);
        apiCalls++;
      }

      // 4. Risk check every minute
      if (apiCalls % 60 === 0) {
        await this.analyzeRisk(trades);  // $0.02
        apiCalls++;
      }

      // Sleep until next iteration (1 second)
      await new Promise(resolve => setTimeout(resolve, 1000));
    }

    // Close payment channel (single on-chain TX)
    await this.channel.close();

    return {
      duration: durationMs,
      trades: trades.length,
      apiCalls,
      pnl: this.calculatePnL(trades),
      costs: {
        api: apiCalls * 0.0001,  // Approximate
        gas: 0.0002,              // 2 TX (open + close channel)
      },
    };
  }

  private async getPriceUpdate(): Promise<number> {
    // Payment channel automatically updates state (off-chain)
    // No user prompt, no on-chain TX, <10ms latency
    const response = await this.browser.fetch('https://price.jup.ag/v1/price', {
      method: 'GET',
      headers: {
        'X-Payment-Channel': this.channel.id,
      },
    });

    // Browser handles x402 payment via channel state update
    return (await response.json()).data.price;
  }

  private async executeTrade(signal: TradeSignal): Promise<Trade> {
    // Jupiter aggregator API with x402
    const response = await this.browser.fetch('https://quote-api.jup.ag/v6/quote', {
      method: 'POST',
      body: JSON.stringify({
        inputMint: signal.fromToken,
        outputMint: signal.toToken,
        amount: signal.amount,
        slippageBps: 50,
      }),
    });

    const quote = await response.json();

    // Execute swap (agent signs with 1-of-3 key share)
    const txSignature = await this.browser.wallet.signAndSend(
      quote.swapTransaction
    );

    return {
      signature: txSignature,
      amount: signal.amount,
      price: signal.price,
      timestamp: Date.now(),
    };
  }

  private calculateSignal(price: number, depth: OrderBook): TradeSignal {
    // Market-making logic: place orders at bid/ask
    const spread = depth.asks[0].price - depth.bids[0].price;

    if (spread > 0.001) {  // 0.1% spread = profitable
      return {
        shouldTrade: true,
        fromToken: 'USDC',
        toToken: 'SOL',
        amount: 1_000_000,  // 1 USDC
        price,
      };
    }

    return { shouldTrade: false };
  }

  private calculatePnL(trades: Trade[]): number {
    // Sum up spread capture from market-making
    return trades.reduce((pnl, trade) => {
      return pnl + (trade.sellPrice - trade.buyPrice) * trade.amount;
    }, 0);
  }
}

// Usage
const agent = new TradingAgent(browser, spendingPolicy);

const report = await agent.runMarketMakingSession(
  10 * 60 * 1000  // 10 minutes
);

console.log(`Session complete:
  Trades: ${report.trades}
  API Calls: ${report.apiCalls}
  PnL: $${report.pnl.toFixed(2)}
  API Cost: $${report.costs.api.toFixed(4)}
  Gas Cost: $${report.costs.gas.toFixed(4)}
  Net Profit: $${(report.pnl - report.costs.api - report.costs.gas).toFixed(2)}
`);

Agent Persona 3: Content Creator Agent

Workflow Overview

A content creator agent generates articles, social media posts, and multimedia content by autonomously researching, writing, and optimizing for engagement. Typical use case: a marketing team that needs 50 blog posts per month.

┌──────────────────────────────────────────────────────────┐
│         CONTENT CREATOR: SINGLE ARTICLE WORKFLOW         │
├──────────────────────────────────────────────────────────┤
│                                                          │
│  Topic: "How threshold signatures secure MPC wallets"    │
│                                                          │
│  1. Research Phase:                                      │
│     ┌──────────────┐                                     │
│     │ Web Search   │  ←  5 searches (Google, Bing)      │
│     └──────────────┘     $0.01 × 5 = $0.05              │
│                                                          │
│     ┌──────────────┐                                     │
│     │ Content      │  ←  10 articles accessed           │
│     │ Access       │     $0.10 × 10 = $1.00             │
│     └──────────────┘                                     │
│                                                          │
│  2. Writing Phase:                                       │
│     ┌──────────────┐                                     │
│     │ GPT-4 Draft  │  ←  2000-word article              │
│     │              │     (50k tokens in, 2k out)        │
│     └──────────────┘     $0.10 + $0.02 = $0.12          │
│                                                          │
│  3. Optimization Phase:                                  │
│     ┌──────────────┐                                     │
│     │ SEO Analysis │  ←  Keyword optimization           │
│     └──────────────┘     $0.20                          │
│                                                          │
│     ┌──────────────┐                                     │
│     │ Plagiarism   │  ←  Originality check              │
│     │ Check        │     $0.15                          │
│     └──────────────┘                                     │
│                                                          │
│  4. Media Generation:                                    │
│     ┌──────────────┐                                     │
│     │ DALL-E 3     │  ←  Featured image                 │
│     │ Image        │     $0.04 per image                │
│     └──────────────┘                                     │
│                                                          │
│  Total: $1.56 per article                                │
│  Time: 5 minutes (vs. 2 hours human writing)            │
│                                                          │
└──────────────────────────────────────────────────────────┘

Content Creator Agent: Single Article

Cost Breakdown

ServiceUsageCost/UnitTotal
Web Search APIs5 searches$0.01$0.05
Content Access (x402 paywalls)10 articles$0.10$1.00
GPT-4 Draft Generation50k in, 2k out$0.002/1k, $0.01/1k$0.12
SEO Analysis API1 report$0.20$0.20
Plagiarism Check1 scan$0.15$0.15
DALL-E 3 Image1 image (1024×1024)$0.04$0.04

Total: $1.56 per article

ROI Analysis

Compare agent-generated content to human writer costs:

  • Human writer: $50/hour × 2 hours = $100 per article
  • Agent cost: $1.56 per article
  • Human review: 10 minutes × $50/hour = $8.33
  • Total agent-assisted: $9.89
  • Savings: $90.11 per article (90% reduction)

For a content marketing team producing 50 articles/month:

  • Human-only: $5,000/month
  • Agent-assisted: $494.50/month
  • Annual savings: $54,066

Code Example

content-agent.ts
import { AgenticBrowser } from '@agentic-browser/sdk';

class ContentCreatorAgent {
  constructor(
    private browser: AgenticBrowser,
    private spendingPolicy: SpendingPolicy
  ) {
    this.spendingPolicy.setArticleLimit(5_00);  // $5 per article
  }

  async createArticle(topic: string, wordCount: number): Promise<Article> {
    console.log(`Creating article: "${topic}" (${wordCount} words)`);

    // Step 1: Research topic
    const research = await this.researchTopic(topic);
    // Cost: ~$1.05 (searches + content access)

    // Step 2: Generate draft
    const draft = await this.generateDraft(topic, research, wordCount);
    // Cost: ~$0.12 (GPT-4)

    // Step 3: Optimize for SEO
    const optimized = await this.optimizeSEO(draft);
    // Cost: $0.20

    // Step 4: Check originality
    const plagiarismScore = await this.checkPlagiarism(optimized);
    // Cost: $0.15

    // Step 5: Generate featured image
    const featuredImage = await this.generateImage(topic);
    // Cost: $0.04

    return {
      topic,
      content: optimized,
      wordCount: optimized.split(' ').length,
      seo: {
        keywords: optimized.keywords,
        readability: optimized.readabilityScore,
      },
      plagiarismScore,
      featuredImage,
      totalCost: 1.56,
      generationTime: '5 minutes',
    };
  }

  private async researchTopic(topic: string): Promise<Research> {
    // Search multiple sources
    const searches = await Promise.all([
      this.searchGoogle(topic),
      this.searchBing(topic),
      this.searchDuckDuckGo(topic),
    ]);

    const urls = searches.flat().slice(0, 10);

    // Access full articles (x402 micropayments bypass paywalls)
    const articles = await Promise.all(
      urls.map(url => this.fetchArticle(url))
    );

    return {
      sources: articles.length,
      content: articles.map(a => a.text).join('\n\n'),
      references: articles.map(a => ({ title: a.title, url: a.url })),
    };
  }

  private async fetchArticle(url: string): Promise<ArticleContent> {
    // Many news sites/blogs now support x402 micropayments
    // Instead of monthly subscription, pay $0.10 per article
    const response = await this.browser.fetch(url);

    // If server returns 402 Payment Required:
    // Browser automatically pays $0.10 via x402, receives content
    return {
      url,
      title: response.headers.get('X-Article-Title'),
      text: await response.text(),
    };
  }

  private async generateDraft(
    topic: string,
    research: Research,
    wordCount: number
  ): Promise<string> {
    // OpenRouter x402 API with cost-basis pricing
    const response = await this.browser.fetch('https://openrouter.ai/api/v1/chat', {
      method: 'POST',
      body: JSON.stringify({
        model: 'openai/gpt-4-turbo',
        messages: [{
          role: 'system',
          content: 'You are an expert technical writer specializing in blockchain and cryptography.'
        }, {
          role: 'user',
          content: `Write a ${wordCount}-word article on "${topic}" using these sources:\n\n${research.content.slice(0, 20000)}`
        }],
        max_tokens: Math.ceil(wordCount * 1.5),
      }),
    });

    return (await response.json()).choices[0].message.content;
  }

  private async optimizeSEO(content: string): Promise<OptimizedContent> {
    // SEO analysis API with x402
    const response = await this.browser.fetch('https://seo-api.example.com/optimize', {
      method: 'POST',
      body: JSON.stringify({ content }),
    });

    return response.json();
  }

  private async checkPlagiarism(content: string): Promise<number> {
    // Plagiarism detection with x402 ($0.15 per scan)
    const response = await this.browser.fetch('https://plagiarism.example.com/check', {
      method: 'POST',
      body: JSON.stringify({ text: content }),
    });

    return (await response.json()).originalityScore;
  }

  private async generateImage(topic: string): Promise<string> {
    // DALL-E 3 via OpenRouter x402
    const response = await this.browser.fetch('https://openrouter.ai/api/v1/images/generate', {
      method: 'POST',
      body: JSON.stringify({
        model: 'openai/dall-e-3',
        prompt: `Professional featured image for article about: ${topic}`,
        size: '1024x1024',
      }),
    });

    return (await response.json()).data[0].url;
  }
}

// Usage
const agent = new ContentCreatorAgent(browser, spendingPolicy);

const article = await agent.createArticle(
  'How threshold signatures secure MPC wallets',
  2000  // word count
);

console.log(`Article generated:
  Title: ${article.topic}
  Words: ${article.wordCount}
  SEO Score: ${article.seo.readability}/100
  Originality: ${article.plagiarismScore}%
  Cost: $${article.totalCost}
  Time: ${article.generationTime}
`);

Comparative Economics

The following table compares the three agent personas across key metrics:

MetricResearch AgentTrading AgentContent Agent
Session Duration3 minutes10 minutes5 minutes
API Calls3069520
API Cost$10.60$0.57$1.56
Gas Cost$0.003$0.0002$0.002
Revenue/Value$200 (labor saved)$15.40 (PnL)$100 (labor saved)
ROI18.9x27.0x64.1x
Payment Channel?No (low frequency)Yes (essential)No (low frequency)

Why x402 Enables These Use Cases

These agent workflows would be impossible without x402's unique properties:

1. Sub-Cent Micropayments

Traditional payment rails (credit cards, PayPal) have fixed fees of $0.30+ per transaction. For a $0.0001 API call, that's 3000x overhead. x402 enables true micropayments with near-zero fixed costs.

2. Machine-to-Machine Commerce

Agents don't have credit cards or bank accounts. x402 uses cryptographic wallets and HTTP-native payment semantics—no account registration, no OAuth flows, no human intervention.

3. Spending Controls

Threshold signatures (2-of-3) allow users to set cryptographic spending limits. Unlike API keys (revocable but not rate-limited), threshold policies enforce $X per hour limits at the signature level. If an agent is compromised, it cannot exceed user-defined thresholds.

4. Self-Custody

The agent holds 1-of-3 key shares. Users retain ultimate control. This aligns with the crypto-native principle: "Not your keys, not your crypto." Custodial wallets (Coinbase Wallet API, etc.) defeat the purpose of autonomous agents—users should not trust centralized services with agent spending authority.

Summary

The agentic browser with x402 support transforms economic models for autonomous agents:

  • Research Agent: $10.60 per report vs. $200 human labor (18.9x ROI)
  • Trading Agent: $0.57 per session for $15.40 profit (27.0x ROI)
  • Content Agent: $1.56 per article vs. $100 human labor (64.1x ROI)

These use cases are only viable because x402 enables:

  1. HTTP-native micropayments (no credit card overhead)
  2. Payment channels for high-frequency operations
  3. Threshold signature wallets for cryptographic spending controls
  4. Self-custody model (agents hold 1-of-3 key share)

As x402 adoption accelerates (10,000% MoM growth in October 2025), the economic viability of autonomous agents will expand dramatically. The agentic browser positions users at the forefront of this shift, maintaining control while unlocking agent capabilities that were previously uneconomical or technically infeasible.