Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →In one developer’s 2026 comparison, all three agent frameworks passed a strict four-key JSON task in nine reported runs, but approval handling and trace reconstruction differed sharply. LangGraph suspended every tested approval workflow; Strands had a duplicate simulated publish call and three empty final outputs across the reported run set; and CrewAI reached 131 model calls after repeated rejection. These are observations from one setup, not a general framework ranking or a production reliability verdict.
The comparison was reported by sunnydachs in a September 21, 2026 DEV Community article. It asked how Strands, LangGraph, and CrewAI handled approval, audit-trace reconstruction, and strict output requirements when given the same agent task.
As an Amazon Associate I earn from qualifying purchases.
The author reports 45 runs in total: 18 approval-gate runs, 36 audit-trail runs analyzed, and nine structured-output runs. These are the author’s experiment counts, not 45 runs for each framework; the findings should be read as setup-specific observations.
Free tools Windows power users keep installed
One-click scans. No signup required.
What the experiment found
| Test | Strands | LangGraph | CrewAI |
|---|---|---|---|
| Approval behavior | Correct action order in 3 of 3 reported runs; one duplicate simulated publish call. No publish after rejection. | Suspended in all 6 reported runs and resumed through a checkpointer; rejection routed away from publish. | Approval took one call and rejection took two in the described test; repeated identical rejection led to 131 LLM calls. |
| Trace reconstruction: rationale / tool order / arguments | 100% / 100% / 100% | 100% / 0% / 0% | 100% / 50% / 50% |
| Four-key JSON task | Met the tested requirement in all 9 reported runs; validation averaged two calls, with four revisions in one run. | Met the tested requirement in all 9 reported runs. | Met the tested requirement in all 9 reported runs. |
| Successful exit with an empty final output | Three instances across the 45-run set. | None reported. | None reported. |
The percentages refer to whether the author could recover specific facts from traces in this particular setup. They do not measure overall audit quality or establish regulatory compliance. All counts and percentages are the author’s measurements, not independently replicated results.
#1 Best Overall
- High-Performance AI Processor：The MS-02 Ultra features an Intel Core Ultra 9 285HX (24C/24T, up to 5.5 GHz, 13 TOPS NPU), delivering fast and efficient performance for AI inference, algorithm development, and media workloads. A PCIe x16 expansion slot supports desktop-class GPU upgrades for advanced model training and accelerated computing tasks. It's ideal for creators, engineers, and teams handling intensive parallel workloads.
- 4 × M.2 PCIe 4.0 + 4 × DDR5 SODIMM slots:Four DDR5 SODIMM slots support up to 256 GB of memory, while ECC helps maintain data integrity in mission-critical environments. Four PCIe 4.0 M.2 slots support up to 24 TB of storage, supporting RAID 0/1/5/10, combining high-speed performance with data protection. It allows for the creation of independent scratch disks, media libraries, and project drives, providing high-throughput for production workflows.
- PCIe & USB 4.0 v2: Up to three PCIe slots can be equipped, including a dual-slot x16 GPU. The main slot supports PCIe 5.0, meeting the needs of high-bandwidth creative and computing workloads. USB 4.0 v2 (80Gbps) supports high-bandwidth external storage and displays.
- Ultra-fast Networking: Wi-Fi 7 further enhances wireless performance with next-generation speeds and low-latency stability. Intelligent bandwidth switching optimizes throughput in different network environments, ensuring optimal performance for enterprise or local networks. Dual 25GbE ports (providing up to approximately 3.125 GB/s bandwidth, about 25 times faster than traditional 1GbE), enabling seamless large-scale file transfers and parallel computing. 10GbE and 2.5GbE ports, with support for Intel vPro technology, ensure enterprise-grade remote management and deployment flexibility.
- Server-grade thermal architecture: Utilizing a dedicated CPU/GPU airflow design, equipped with a 6-pipe dual-fan cooler, it maintains stable performance even under sustained loads, delivering up to 140W Turbo power while maintaining a 100W TDP, and operating with noise levels as low as 36 dB. An integrated 350W power supply ensures stable and reliable output for demanding computing tasks and fully loaded extended configurations.
Approval: workflow enforcement versus a request to the model
The approval task created a news digest, sought reviewer approval, and only then attempted a simulated publish action. The difference in the reported results is useful when deciding what to test: is approval a hard workflow boundary, or does the design rely on the model following an instruction?
Strands: prompt-led approval, with a duplicate-action failure
In the reported Strands implementation, the prompt asked the model to seek reviewer approval. The author says the correct order was followed in all three runs and no publish action occurred after rejection. However, one run called the simulated publish action twice. That matters because a workflow can obey the approval sequence and still repeat a side effect. The experiment does not establish how a real publishing system would handle duplicate requests.
LangGraph: explicit pause and resume
The LangGraph workflow used interrupt() to pause and Command(resume=...) to continue, with a checkpointer supporting resumption. It suspended in all six reported approval runs, and rejection routed away from publishing. That is evidence about the tested graph design, not a claim that every LangGraph application automatically enforces approval; the enforcement depends on how the workflow is built.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →CrewAI: feedback can become an unbounded loop
The CrewAI test used Task(human_input=True) to request console feedback after the task. Approval took one call and rejection two in the described test. When the author repeated the same rejection, the run reached 131 LLM calls. This is a reason to test rejection limits, exit conditions, and what happens when feedback does not change—not a measured general call count for CrewAI.
Audit traces: recoverability depended on implementation
The author scored whether seven audit-relevant facts could be reconstructed, including rationale, tool order, tool arguments, and model identity. In trace-only reconstruction, rationale was recoverable in 100% of runs for each framework. Tool order and arguments were recoverable in 100% of Strands runs, 0% of LangGraph runs, and 50% of CrewAI runs.
The author attributes LangGraph’s missing tool-order and argument evidence to tools being executed in code rather than sent as model tool calls in this implementation. This distinction is important: a trace may explain the model’s reasoning while omitting the sequence or parameters of actions performed by the surrounding program. The result is not proof that LangGraph cannot produce such evidence; it shows that framework-level design and instrumentation affect what an auditor can reconstruct.
Rank #3
- Professional AI & Creator Workstation: AMD Radeon AI PRO R9700 GPU with 32GB GDDR6 is engineered for AI development, professional content creation, and compute-intensive workloads.
- Massive 32GB Memory Capacity: 32GB of GDDR6 memory on a 256-bit bus provides ample bandwidth for large AI models, 8K video editing, and complex 3D rendering.
- Advanced RDNA 4 with AI Accelerators: 64 Compute Units with 3rd Gen Ray Tracing and dedicated 2nd Gen AI Accelerators for groundbreaking AI performance and visual computing.
- Professional Blower Cooling: Efficient single blower design exhausts heat directly out of the chassis, ideal for multi-GPU workstation and server configurations.
- Enterprise-Grade Thermal Solution: Vapor chamber heatsink with industrial Honeywell PTM7950 thermal interface material ensures reliable cooling under sustained professional loads.
For a real evaluation, inspect the actual event records your application retains. Check whether they preserve the approval decision, action order, tool inputs, model identity, retries, and outcome in a form that an authorized reviewer can retrieve. A percentage from this experiment is not a substitute for validating your own retention and access requirements.
Recommended Free Tools
Strict output: all three passed this narrow test
The structured-output task required JSON with exactly four keys: summary, word_count, topics, and publish_ready. The author reports that all three frameworks met the requirement in all nine structured-output runs, and the reported word count matched the summary length each time. Strands’ validation loop averaged two calls and required four revisions in one run.
This finding applies to that schema and task. It does not establish that any framework will always produce valid output for a different schema, longer prompt, changed model, or downstream parser. Test the exact schema and failure handling your application needs.
Rank #4
- FAST RUNS IN THE FAMILY — The 16-inch MacBook Pro with the M5 Pro or M5 Max chip brings next-generation speed and powerful on-device AI to personal, professional, and creative tasks. With all-day battery life, double the starting storage,* and a breathtaking Liquid Retina XDR display, it’s pro in every way.*
- BUCKLE UP — Along with a next-generation CPU, faster unified memory, and up to 2x faster SSD storage,* M5 Pro and M5 Max feature a more powerful GPU with a Neural Accelerator built into each core, delivering faster AI performance and on-device training capabilities. So you can blaze through demanding workloads at mind-bending speeds.
- BUILT FOR AI — Apple silicon, and every major component that powers it, is designed to run demanding on-device AI workloads like LLM inference and training. And Apple Intelligence helps you write, express yourself, and get things done effortlessly with groundbreaking privacy protections at every step.*
- ALL-DAY BATTERY LIFE — MacBook Pro delivers the same exceptional performance whether it’s running on battery or plugged in.*
- MACOS RUNS APPS FAST — All your go-to apps run lightning fast in macOS, including built-in apps like FaceTime and Messages. Plus, built-in virus protection and free software updates help keep your Mac running smoothly and securely.
Empty final results can hide behind successful exits
Across the reported 45-run set, Strands had three cases where the process exited successfully but the final output was empty. The author says the completed result could be recovered from the trace’s tool-call argument. That may help investigation, but it does not prevent a downstream consumer from receiving no deliverable. LangGraph and CrewAI had no such cases in these reported tests.
For applications that hand results to another service, treat a successful process exit and a usable result as separate conditions. Validate that the final field exists and is nonempty before accepting a run, and decide what recovery path to take when the trace contains information that the normal output omitted.
How to use these results in a framework evaluation
The benchmark is most useful as a source of test cases, not as a winner announcement. For your own implementation, measure the behavior that matters at the boundary between the agent and its users or systems:
Best Value
- 【High-Performance APU】The MS-S1 MAX features an AMD Ryzen AI Max+ 395 APU, integrating a Zen 5 architecture CPU (up to 5.1GHz, 16C/32T, 64M L3 Cache), an RDNA 3.5 GPU, and an NPU (50 TOPS). The total system output is 126 TOPS. It provides powerful parallel computing capabilities for demanding AI workflows. It is ideal for running local LLMs, multimodal models, and computationally intensive tasks
- 【128GB UMA Memory】Equipped with up to 128GB of LPDDR5x-8000MT/s unified memory, it enables the CPU and GPU to access a shared, high-bandwidth memory pool with extremely low latency. Ideal for large-scale AI inference, 3D workloads, and complex timelines in video editing. It eliminates traditional VRAM bottlenecks, ensuring smoother data transfer during high-intensity computations. The UMA design maximizes performance stability under high loads
- 【Flexible Expansion】The MS-S1 MAX features USB4 V2 (up to 80Gbps), dual 10GbE LAN, HDMI 2.1 (up to 8K60), a full-length PCIe x16 expansion slot, and dual M.2 slots supporting up to 16TB RAID 0/1. Wi-Fi 7 provides stronger signal coverage and a more stable wireless experience. The slide-out design facilitates upgrades and maintenance. It easily adapts to personal, studio, or rack-mount enterprise environments
- 【High-Efficiency Cooling System】Utilizing an aerospace-grade aluminum alloy chassis, copper base plate, six heat pipes, dual turbine fans, and advanced PCM thermal conductive material, it maintains stable cooling performance even under continuous load. This system supports 130W continuous power and 160W peak power operation, with a built-in 320W power supply. It boasts multiple global certifications including CCC, FCC, UL, CE, and UKCA, ensuring stable and reliable operation in various environments
- 【Cluster Design】Two MS-S1 MAX units can be configured as a dual-unit cluster to run a large 235B Q4 model locally, achieving an output speed of 10.87 tok/s. Supporting 2U rack deployment, multiple MS-S1 MAX units can be cascaded into a distributed cluster to create a high-efficiency AI computing center. A cluster of four MS-S1 MAX units successfully ran a DeepSeek-R1 671B Q4 large model. A reserved cluster power-on interface allows for unified start-up and shutdown
- Approval: Confirm whether a rejection structurally prevents the action, rather than merely asking the model not to proceed.
- Repeated feedback: Set and test a bounded retry or call policy for unchanged or contradictory reviewer responses.
- Side effects: Check whether retries can repeat an action, and whether the receiving system safely handles duplicate requests.
- Trace evidence: Verify that reviewers can retrieve rationale, tool sequence, arguments, model identity, and outcomes from retained records.
- Output contract: Validate exact keys, types, and required values with the parser or schema check your application will use.
- Completion: Reject successful process exits that have no usable final result.
- Recovery: Test resumption after an actual process crash, not only a normal pause and resume.
These are evaluation questions prompted by the reported failure modes. The comparison did not measure every one of them.
Limits that change how much to infer
The author says the tests used the same recorder proxy, model, and tools, but does not name the model provider or framework versions. The setup used a scripted human, had no real notification or review UI, and simulated the destructive publish action. Each test cell had three runs, as the author summarized: “One model, 3 runs per cell – directional, not a definitive ranking.”
The associated agent-framework-showdown GitHub repository is described as containing commands for the experiments. The results presented here are the author’s report; the commands and findings have not been independently verified. The small, single-setup comparison can surface failure modes to probe, but it cannot establish population-level failure rates, production reliability, or a universal best choice.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteQuick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
