Technical guides for building reliable AI agents.

Tangle builds isolated runtimes, model routing, browser automation, and evaluation tools for AI agents. These guides explain how to use those pieces, how we measure them, and where their evidence stops.

New here? Start with Tangle Re-Introduction for the protocol or Agent Intent Infrastructure for agent services.

Topics

Choose a learning path

Each series follows one technical subject from first principles through implementation, measurement, and limits.

An editorial still life about describing and running an agent task

Agent Intent Infrastructure

Learn how software discovers a Tangle service, runs a bounded task, pays per request, and keeps evidence. A Blueprint is a reusable service definition, a Job is one callable unit, and an operator runs that Job.

An editorial still life about the software that runs an agent

Agent Runtime Infrastructure

An agent runtime is the software that gives a model its tools, files, limits, and record of work. Follow those settings into isolated workers, then see how the runtime handles budgets, failures, output checks, and traces.

An editorial still life about a runnable service blueprint

Blueprint Agent

Explore a browser-based coding workspace where an agent works inside an isolated Sandbox—a disposable environment where it writes and runs code—and completes small quests checked by builds, tests, browser behavior, runtime state, or deployment evidence.

An editorial still life about a developer kit for runnable services

Blueprint SDK Guide: Run Rust Services on Tangle

Tangle is a network where independent operators run services. The Blueprint SDK is a Rust toolkit for packaging, testing, and operating those services: a Blueprint is a reusable service definition, a Job is one callable unit of work, and a Service is one live configured instance run by an operator.

An editorial still life about an agent operating a web browser

Browser Agent

A browser agent drives real Chromium from a natural-language goal and returns its observations, actions, screenshots, and result. A fixture is the controlled test data prepared before a run; provider state is the wallet’s current account, network, and connection.

An editorial still life about preparing a complex tax return with software

Building an AI Tax Agent

This proposed preparation workflow turns source documents into linked facts, draft forms, and workpapers—the calculations and records supporting a return—inside a review packet that a person inspects before filing.

An editorial still life about reviewing code and producing evidence

Code Auditor

This series describes agent-assisted code review in which scanner alerts enter an isolated workspace for reachability, impact, and reproduction tests. An alert is a candidate; a finding is a claim another reviewer can reproduce.

An editorial still life about operators running services on a network

Tangle Protocol

Tangle coordinates services run by independent operators. A Blueprint is the service template, a Service is one configured instance, and a Job is one unit of work; the actual computation runs off-chain.

An editorial still life about building a service on Tangle

Tangle Re-Introduction

Start from first principles: Tangle coordinates operator-run services defined as Blueprints, deployed as Services, and called through Jobs. The series moves from that architecture to verification, deployment, inference, sandboxes, and hardware-isolation evidence.

An editorial still life about measuring an agent system

The Instrument Problem

An agent evaluation is a repeatable test used to compare systems or decide whether a change should ship. This series examines four ways that an otherwise precise result can answer the wrong question: an incomplete benchmark, a weak behavioral signal, a ranking rule that disagrees with release policy, and a cost total that omits workers.

An editorial still life about how an agent system improves

The Self-Improving Stack

Map the parts an agent team can change—prompts, skills, runtime, traces, evaluations, model training, and release rules—and the evidence required before promoting a candidate change.

A precision balance weighs one blue glass token against five teal tokens beside a brass selection ring.

When Structure Pays

An agent system can spend extra computation in two ways: generate several candidates and choose among them, or add a parent process that delegates to workers. These articles compare both choices against a simpler one-agent baseline.

An editorial still life about taking a paid agent service to production

x402 Production Runway

x402 turns an HTTP 402 payment challenge into signed payment terms and a retried request. On Tangle, a Blueprint is the typed service contract: each paid request becomes one Job declared by it, facilitators verify and settle payment, and operators execute the work.

All technical articles

Browse every article by publication date, or use the topic filter above to narrow the list.

Why AI Agent Token Budgets Need a Measured Runtime Floor

An AI agent token budget can be impossible before work starts. A measured runtime floor lets a parent reject unsafe child budgets and preserve unknowns.

Why AI Agent Delivery Checks Must Run After Worker Failure

An AI agent can write a valid artifact and still exceed its budget. Separate delivery, resource compliance, and failure propagation so each result stays true.

Agent Runtime Trace Writer-Reader Contracts

An agent trace reader depends on a stable writer contract. Publish the layout, move state to .agent, and classify every backend at compile time.

AI Agent Cost Accounting: Count Every Child Worker

An AI agent run can look cheap when its ledger counts only the driver; reconcile child-worker tokens and mark missing usage as unknown before comparing systems.

AI Agent Evaluation: Output Decay Is Weak Evidence

Three shrinking model responses are weak evidence of agent decay; five calls plus a 60% endpoint make this behavioral check harder to fool today.

AI Agent Evaluation: Match Selection to the Release Rule

An AI evaluation can pick the wrong winner when its average score disagrees with the fail-closed rule used to release an agent candidate safely.

When Does Best-of-Five Sampling Pay for AI Code Generation?

Best-of-five sampling improves AI code generation when every candidate faces the same deterministic check; learn the measured lift, cost, limits, and rule.

Supervisor vs. Solo AI Agents: Measure Coordination Cost

A paired replay found solo 7/10 versus supervisor 4/10, p=.25, with 2.55× tokens and 2.42× wall time; learn the result's limits and fair test.

AI Agent Profile: Make Settings Deliverable

An AI agent profile lists the model, tools, files, and policies a run needs. A delivery contract rejects unsupported settings before the worker starts.

AI Agent Runtime Architecture: When to Remove a Path

AI agent runtime architecture can remove a second execution path when the shared route preserves requested settings, controls, errors, and trace records.

AI Agent Observability: Make the Work Visible

AI agent observability should expose live progress and tool requests, while leaving completion and duration unknown when the remote stream omits them.

AI Coding Agent Benchmark: What CodeTraceBench Measures

CodeTraceBench, an AI coding-agent benchmark, gave recursive and one-shot analysts nearly the same localization score while recursion cost 5.57 times more.

AI Coding Agent API Integration: Measuring Agent Readiness

Measure AI coding agent API integration with current-contract tasks, hidden execution checks, calibration, denominators, and a public Stripe benchmark board.

Deploy a Paid AI Agent Service: One Traceable Job

A practical path for exposing one paid AI job with discovery, authorization, payment, execution, recovery, and evidence that a buyer can inspect.

AI Agent Runtime Environment: Tools, State, and Proof

An AI agent runtime environment gives a model tools, files, permissions, and records for real work. Follow a toy tax-document review workflow.

Natural Language E2E Testing for Wallet Apps

Natural-language E2E testing for wallet apps lets agents drive browser flows while stopping before destructive signing and preserving evidence.

OpenAI Compatible Routers for Agents

OpenAI-compatible routers keep one request shape while exposing model discovery, routing policy, usage records, and provider or cost changes.

Browser Automation for AI Agents: Evidence and Safe Stops

Browser automation for AI agents needs page state, screenshots, recovery, and a clear stop condition so a reviewer can tell what happened.

AI Agent Sandbox: Build a Controlled Agent Workspace

An AI agent sandbox gives a software agent isolated files, processes, network rules, and reviewable output. Learn how to test the workspace with a real task.

AI Accountant Workpapers for Complex Tax Returns

Trace each reported value to a source-linked workpaper, surface conflicting facts, and prepare the questions a reviewer must resolve before filing.

AI Browser Testing With Evidence Traces

Design the base record for one browser run: initial state, ordered actions, screenshots, a final assertion, and a failure another reviewer can replay.

AI Code Audit: From Scanner Alert to Reproducible Evidence

An AI code audit turns scanner alerts into findings by testing reachability in an isolated environment, recording impact, and preserving rejected candidates.

AI Coding Assistant With Deployment Evidence

AI coding assistants become useful for onboarding when they turn a brief into running code, checked tasks, deployment evidence, and a reviewable trace.

AI Dev Container For Production Agents

An AI dev container needs isolation, command execution, durable sessions, trace export, and explicit failure handling before an agent touches a real repository.

AI E2E Testing For Browser Flows

Follow a full browser journey across routes and backend state, using controlled test data and a final condition that proves the user's task finished.

AI Security Audit: Make Every Finding Reproducible

An AI security audit becomes useful when each finding carries its code location, exploit path, command output, severity reasoning, and fix.

AI Service Marketplace With Crypto Payments: Job to Result

An AI service marketplace with crypto payments lets agents discover a defined job, pay per request, validate results, and recover from failure.

AI Tax Filing Software for Complex Returns

Bind each draft tax form to source-document versions, corrections, unresolved questions, and the exact package approved before electronic filing.

AI Tax Preparation for Complex Returns

Build source-backed workpapers with the calculations and records supporting a return, then connect them to draft forms and review questions.

AI Vulnerability Scanner vs. Agent Audit: Choose by the Risk

An AI vulnerability scanner finds possible issues across code; an agent audit tests reachability, validates impact, removes duplicates, and explains the fix.

Anonymous LLM Usage: Shielded Payment Limits

Anonymous LLM usage depends on the observer and the request data in question, while shielded payments address only one part of the payment and logging path.

Automated Smart Contract Audit: Prove High-Severity Findings

An automated smart contract audit should validate high-severity findings with a test, simulation, trace, or proof of concept before assigning severity.

Automated Tax Filing With Review Before Submit

Bind human approval to one evidence-backed return version and pause submission whenever a source document, calculation, or form changes afterward.

Blueprint Protocol: Define an Operator-Run AI Service

Define a Blueprint protocol service with a typed job, reproducible runtime, payment rule, evidence record, and operator runbook.

Blueprint SDK Deployment: Run an Operator Job

Use Tangle's Blueprint SDK to turn a Rust service into an operator-run job with a clear contract, evidence from a deployment on a test network, monitoring, and rollback.

AI Browser Automation Needs An Evidence Loop

Decide when a browser agent should observe again, retry a changed page, stop, or fail without letting a moved interface become a false pass.

CFC Tax Filing Software for Form 5471

Build the ownership timeline behind Form 5471, link foreign records to each schedule, and stop before filing when a required fact is missing.

Complex Tax Situations Software For Founder Returns

Connect company ownership, K-1s, equity, crypto, foreign reporting, and state records in one source map before drafting a founder's return.

Controlled Foreign Corporation Taxes And Form 5471

Use an ownership timeline to choose Form 5471 filing categories, then link foreign records and reviewable workpapers to each required schedule.

Crypto Hackathon Platform: Judge Builds With Code Evidence

Compare submissions with separate proof of code, runtime execution, wallet state, and deployment; use task-specific checks to judge correctness.

Crypto Tax Software 2026 for DeFi, Staking, and Wallets

Reconcile broker forms with wallet history, DeFi, staking, transfers, and missing basis before drafting Form 8949 transactions.

Decentralized Compute Protocol: How Tangle Runs Services

An operator-run compute protocol turns a defined job into a service with a visible payment path, result record, and failure policy.

DeFi Wallet Testing With Browser Agents

Separate a DeFi flow into the dapp page, wallet prompt, chain state, and approval boundary using controlled test data and a disposable wallet.

Developer Onboarding Platform With Code-Verified Quests

Separate developers who shipped a working integration from those who only opened docs by checking a build, API call, and reviewable run evidence.

Developer Quest Platform With Code Verification

Define one technical behavior, run its build, test, or browser check in an isolated workspace, and return the exact failure when it breaks.

How AI Agents Discover Products

AI agents discover products through stable URLs, scoped packages, safe calls, OpenAPI files, manifests, and READMEs they can verify.

K-1 Tax Filing For Multiple Entities

Index every K-1 and attachment, link each to basis, passive-loss, and state workpapers, and carry late corrections through the draft.

LLM Sandbox Environment For Agent Runs

An LLM sandbox environment isolates tools, records side effects, survives reconnects, and gives reviewers enough evidence to approve or reject an agent run.

MetaMask Automated Testing For Wallet Flows

MetaMask automated testing should inspect account access, chain changes, signature prompts, rejection recovery, and transaction state with browser evidence.

Natural Language Test Automation That Leaves Proof

Turn a plain-English test goal into setup data, browser checkpoints, and a deterministic final assertion instead of trusting the agent's pass report.

Operator Staking for AI Blueprints: What Stake Proves

Operator staking gives AI Blueprints economic accountability, but stake is not a service-quality score or proof that a model answer is correct.

S Corp Tax Software For Basis And K-1s

Reconcile Form 1120-S and each K-1 with shareholder stock basis, debt basis, distributions, payroll, and suspended-loss history.

Tangle Browser Agent vs Browserbase and Browser Use

Compare Tangle Browser Agent, Browserbase, and Browser Use by task evidence, browser sessions, workspace control, and API surface.

Tangle Sandbox vs Daytona and Modal

Compare Tangle Sandbox, Daytona, and Modal by work unit: durable agent computer, development sandbox, or serverless job or GPU workload.

Tangle Sandbox vs E2B: Choosing An AI Agent Sandbox

Tangle Sandbox and E2B both isolate agent code, but differ in durable sessions, workspace recovery, templates, and trace-oriented review.

TEE Attestation for AI Services: What the Evidence Proves

TEE attestation can bind an AI service request to approved code on protected hardware, but it cannot prove the answer is correct. Check reports and secrets.

Web3 Developer Tools Need An Agent Workbench

Web3 developer tools need an agent workbench that connects code, wallets, networks, browser checks, and evidence in one build path.

x402 Payments for AI Agents: v2 Safety Guide

Learn how x402 payments for AI agents expose payment requirements, prevent duplicate work, reconcile settlement, and separate receipts from proof of results.

Agent Governance: How to Govern Self-Improving Agents

Agent governance sets owners, authority limits, evidence requirements, approvals, and rollback paths before behavior changes reach users.

Multi-Agent Coordination: Roles Are Not Structure

Multi-agent coordination needs role contracts, state boundaries, budgets, checks, and traces; compare it with a single-agent baseline at equal cost.

Agent Runtime Topology: When Parallel Work Is Real

Agent runtime topology determines whether prompts produce parallel work, reviews, retries, cancellation, and evidence, or only describe them.

Harness Evolution for AI Agents: When Prompts Plateau

Harness evolution changes the software around an agent when prompts and skills cannot add tools, isolation, checks, traces, or rollback.

Test-Time Compute for Agents: Beat Random at Equal Cost

Test-time compute adds samples, branches, retries, or checks; an execution shape matters only when it beats a simple baseline at equal cost.

Agent Traces: The Evidence an Improving System Needs

Agent traces preserve calls, tool actions, observations, artifacts, costs, and outcomes so the team can diagnose failures, not tune scores alone.

Evaluation Gates: Decide What Agent Changes Ship

Evaluation gates decide whether a change ships by comparing candidate and baseline on protected tasks, deterministic checks, cost, and trace evidence.

Agent Memory: Why Retrieval Is Not Learning

Agent memory becomes learning only when a scoped, evidence-backed write reaches the right future task and evaluation shows better behavior than no memory.

Post-Training Agents: When to Change the Model

Post-training for agents moves improvement from prompts and skills into model weights or adapters, with harder data, release, and rollback decisions.

Optimization Theory for Agent Builders

Optimization theory for agent builders helps choose what to change, compare candidates fairly, and stop noisy gains from becoming production regressions.

Skill Optimization for AI Agents: Durable Procedures

Skill optimization improves reusable agent procedures and their activation rules while keeping model weights fixed.

Prompt Optimization for AI Agents: What It Can Change

Prompt optimization searches instructions, examples, schemas, tool descriptions, and rubrics for better agent behavior while the model and runtime stay fixed.

The Self-Improving Stack: How AI Agents Get Better

The self-improving stack is the set of agent layers that can change, from prompts and skills to runtime, traces, evaluation, model training, and governance.

Tangle Blueprints: How to Choose an Operator-Run Service

Tangle Blueprints are service definitions run by operators. Compare jobs, access, payment, evidence, and failures before choosing an agent service.

DeMo: Communication Reduction for Distributed Training

DeMo, a distributed training method, reports up to 85x less communication per GPU; a service still needs checkpoints, evidence, recovery, and evaluation.

Recursive Self-Aggregation: Test-Time Scaling

Recursive Self-Aggregation uses extra inference calls to combine candidates within a visible budget, with traces, latency, cost, and task evaluation.

Payment-Native Infrastructure for AI Agent Products

Payment-native infrastructure for AI agents: how request-level payment, operator-run execution, evidence, and clear job contracts change what is worth building.

How to Deploy an AI Agent Service With Remote Providers

Deploy an AI agent service with paid HTTP ingress, remote operators, reproducible runtimes, health checks, and rollback for failed jobs.

Secure Code Execution for AI Agents: TEEs and Payment

Secure AI-agent execution uses payment for admission; trusted environments limit access; attestation identifies workloads; result checks test correctness.

x402 Blueprint Production Deployment Checklist

A rollout checklist for taking an x402-enabled Blueprint from dev to staging to mainnet without silent payment or config failures.

Operator Health Monitoring: Heartbeats and Recovery

Diagnose stale heartbeats, expired quotes, and worker failures, then decide which checks must pass before reopening paid traffic.

RFQ Job Quotes on Tangle

Implement signed RFQ fields, expiry, replay protection, result authorization, and recovery when the selected Tangle operator fails.

x402 Operator Economics: Pricing Two Payment Paths

x402 operator economics starts by separating HTTP payments from Tangle service fees, then pricing operator costs, stake shares, payouts, and penalties.

Blueprint Operator Monitoring: Health and Quotes

Choose the runtime, heartbeat, quote, payment, and result signals an operator should expose before a service quietly stops accepting work.

On-Chain Compute Quotes: RFQ and Slashing

Design the quote protocol: bind inputs, operators, and price to a job, restrict result submission, and define evidence and disputes before slashing.

Subscription vs Pay-Per-Request API Pricing

Compare subscription and pay-per-request API pricing across margins, billing work, customer needs, latency, and a hybrid implementation for teams and agents.

Pay-Per-Request API Pricing: Wei and Markup

Pay-per-request API pricing in Tangle explains how wei amounts become accepted-token units, including decimals, markup, and quote freshness.

The x402 Facilitator: Trust, Uptime, and Safer Failure

An x402 facilitator verifies and settles payments; compare hosted, self-hosted, and local options for uptime, trust, and safe failure.

x402 Payments Blueprint: Turn HTTP into a Job

An x402 payment authorizes one Tangle Blueprint job over HTTP; this guide covers settlement, quote expiry, retries, and evidence after execution.

Secure Containers for AI Agents: What TEEs Can Prove

A secure container for AI agents needs a clear data boundary and workload evidence. This guide covers Tangle TEE policy, attestation, secrets, and limits.

AI Agent Infrastructure: Inference and Code Execution

Build AI agent infrastructure on Tangle for model inference and generated code execution, with operator evidence, job boundaries, and failure checks.

How to Build a Tangle Blueprint: Test and Deploy

How to build a Tangle Blueprint from one typed job to a tested operator service, with public SDK commands for local Anvil testing and testnet deployment.

How Decentralized AI Infrastructure Verifies Work

How decentralized AI infrastructure verifies work through result checks, thresholds, hardware attestation, proofs, and task evaluations, with clear limits.

How Blueprints Work

How Blueprints work as reusable Tangle service templates, and how jobs move from public metadata to operator execution, payment, verification, and expiry.

Why Decentralized AI Infrastructure?

Why decentralized AI infrastructure matters when an agent must choose an operator, protect private inputs, and inspect how paid work was executed.