Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minuteSome links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
Amazon, OpenAI and China’s Zhipu (now branded internationally as Z.ai) did not release equivalent products in spring 2025. Amazon previewed Nova Act for browser automation, OpenAI introduced tools for developers building agents and promised an open-weight model, and Zhipu expanded its AutoGLM agent ecosystem. Together, the announcements showed competition moving from chat responses toward systems that can use tools, operate software and run under different distribution and deployment models.
Three announcements, three different layers of the AI market
The developments were grouped together because they addressed the same strategic shift: making AI complete tasks rather than merely generate text. Their commercial positions were different.
| Company | Development | Primary competitive layer |
|---|---|---|
| Amazon | Nova Act research-preview model and SDK | Browser agents, AWS infrastructure and enterprise automation |
| OpenAI | Responses API and agent-building tools in March 2025; open-weight models later that year | Developer platform, model distribution and deployment choice |
| Zhipu/Z.ai | AutoGLM and GLM agent ecosystem | Chinese consumer and developer adoption, autonomous task execution and open models |
The original Computerworld article was published on April 1, 2025, while the underlying announcements were primarily made in March. That timing matters: OpenAI’s open-weight models were planned at the time, not yet available.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallCrashes, 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 minuteComputerworld’s April 1, 2025 report described the contemporaneous developments.
Amazon Nova Act: an early browser-operating agent
What Amazon announced
On March 31, 2025, Amazon introduced Amazon Nova Act as a research preview, alongside an SDK for developers. It was designed to operate websites and browser interfaces: searching, navigating pages, answering questions about a screen and completing structured actions. Amazon described a workflow in which developers break a complex job into smaller browser commands, add detailed instructions and combine those actions with API calls.
This is different from a general-purpose chatbot. Nova Act was aimed at software that can manipulate a web interface, potentially including:
- Website and user-interface testing.
- CRM updates and form completion.
- Reservations, shopping or other structured web workflows.
- Repetitive customer-service and back-office tasks.
At launch, Amazon said U.S.-based customers with an Amazon account could explore Nova through nova.amazon.com. The announcement described Nova Act as an experiment-stage technology, not a mature autonomous employee.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Nova models, Nova Act and Bedrock are not the same thing
Amazon’s Nova foundation models were available through Amazon Bedrock, while Nova Act referred specifically to browser-agent technology and its developer SDK. Bedrock is the broader AWS service for accessing and deploying models. Keeping those products separate helps explain Amazon’s strategy: use an experimental agent capability to attract developers, then move successful workloads into AWS infrastructure and governance.
Amazon’s announcement is available at Amazon’s Nova Act overview.
Where browser agents break
A browser agent can fail when a site changes its layout, presents an ambiguous control, requires multifactor authentication or a captcha, or includes malicious instructions in page content. A timeout can also cause an agent to repeat a transaction. Payments, account changes, deletions and legal or medical submissions therefore need explicit human approval, transaction safeguards and detailed logs.
OpenAI: agent infrastructure first, open weights later
What was available in March 2025
On March 11, 2025, OpenAI announced the Responses API and related tools for developers and enterprises building agents. The package was intended to simplify multi-step tasks and tool use through hosted infrastructure. It was a developer-platform release, not the launch of a new consumer assistant.
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 →Read the announcement at OpenAI’s agent-tools post.
The date correction on open-weight models
In the March-and-April 2025 timeframe, OpenAI’s open-weight model was a future plan. It should not be described as an available model on April 1. OpenAI released gpt-oss-120b and gpt-oss-20b on August 5, 2025.
| Date | OpenAI development |
|---|---|
| March 11, 2025 | Responses API and agent-building tools announced |
| April 1, 2025 | Computerworld reported the planned open-weight release |
| August 5, 2025 | gpt-oss-120b and gpt-oss-20b released |
OpenAI says the models are Apache 2.0 licensed and intended for local desktops, laptops and data centers, with strengths in reasoning, tool use, customization and agentic tasks. Its model information is at the gpt-oss release page and OpenAI’s open-models page.
“Open-weight” is narrower than “open source”
Open-weight access provides model parameters that customers can run or adapt. It does not necessarily provide the training data, complete source code, full data-processing pipeline or a reproducible training recipe. The trade-off is greater deployment control and customization, but also responsibility for hardware, serving, security updates, evaluation and misuse prevention.
Recommended Free Tools
OpenAI’s safety material discusses the additional risks of released weights, including adaptation for harmful purposes: gpt-oss model safety information.
Zhipu and the expansion of China’s agent ecosystem
What the contemporaneous coverage said
Computerworld characterized Zhipu’s offering at the time as a free AI assistant intended to strengthen the company’s position in China. That description should be attributed to the report because the exact product name, rollout scope and consumer terms are not fully established by the cited company material.
What Zhipu’s own history confirms
Zhipu’s official milestone history lists AutoGLM Rumination in March 2025, describing it as an AI agent with deep-research and operational capabilities. The company’s broader family includes GLM foundation models, AutoGLM autonomous agents, ChatGLM, model APIs, search tools and fine-tuning services. The company now uses the international brand Z.ai.
See Zhipu’s company history and its product site. The safest conclusion is that Zhipu was expanding a China-focused assistant and agent ecosystem, rather than releasing a direct equivalent of Nova Act or OpenAI’s API.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Why the launches mattered strategically
The frontier moved from answers to actions
Amazon focused on operating a browser, OpenAI on giving developers agent infrastructure, and Zhipu on consumer and research-oriented autonomous assistance. The central question became whether a system could complete a task reliably, recover from errors and show evidence of what it had done.
Distribution became as important as model quality
- Amazon: AWS, Bedrock, enterprise software and nova.amazon.com.
- OpenAI: APIs, ChatGPT, developer tools and downloadable weights.
- Zhipu/Z.ai: Chinese consumer applications, model APIs and open-model adoption.
For investors and technology buyers, access to users, cloud capacity and developers can matter as much as a benchmark score.
Business models diverged
Hosted APIs monetize usage. Cloud platforms monetize compute, storage, deployment and related services. Open-weight models can accelerate ecosystem adoption while shifting infrastructure and support costs to customers. A free assistant can prioritize user acquisition, but “free” does not automatically mean suitable for business data or regulated work.
Geopolitics added practical constraints
The competition was not simply a two-country contest: Chinese firms compete with one another, and U.S. firms compete intensely among themselves. Buyers must also consider data residency, export controls, local hosting, licensing, government procurement, chip supply and cloud availability.
How businesses should compare the options
| Decision | Usually preferable when | Main trade-off |
|---|---|---|
| Browser agent | No stable API exists and the task is narrow and low risk | Fast prototyping, but fragile and difficult to test exhaustively |
| API-first automation | The process is repeatable and the system exposes documented interfaces | More integration work, generally better determinism |
| Hosted model | Speed, broad capability and simple operations matter most | Provider dependence and data-governance review |
| Open-weight model | Local control, customization or data residency justify operating infrastructure | Customer bears compute, security, evaluation and maintenance |
| China-focused provider | Operations center on China or Chinese-language use cases | Check residency, licensing, support and procurement compatibility |
Questions to answer before deployment
- What is the task? Prefer structured API calls for repeatable production work; reserve screen automation for systems that lack suitable APIs.
- What can go wrong? Define which actions require human approval, especially payments, transfers, deletions and submissions.
- Where does data run? Check processing regions, retention, training-use policies, logs and administrator controls.
- What is the total cost? Include model or session charges, cloud GPUs, engineering, monitoring, security review and human oversight—not just a quoted API rate.
- Can you leave? Assess portability of prompts, tools, workflows, fine-tunes and model weights before committing to a provider.
Risks that should not be hidden by the word “agent”
Browser-agent risks
- Misreading controls or acting on stale page state.
- Prompt injection from hostile webpage content.
- Credential or sensitive-data leakage.
- Duplicate actions after timeouts.
- Breakage after redesigns, pop-ups, captchas or authentication changes.
- Plausible but incomplete results.
Use allow-listed domains, least-privilege credentials, reversible actions, screenshots and tool-call logs. Add deduplication, rollback procedures and human checkpoints for irreversible outcomes.
Open-model risks
Self-hosted or downloadable weights do not remove hallucinations, unsafe fine-tunes, licensing questions, serving vulnerabilities or data leakage through telemetry. Performance can also vary substantially with hardware and the serving stack.
Free-assistant risks
A free product may be geographically limited, consumer-only or missing audit logs, administrator controls, contractual support and predictable uptime. Verify the terms and data handling before placing business or financial information in it.
What this means for technology and finance decision-makers
There was no universal winner in these announcements. Amazon offered a path from browser-agent experimentation into AWS; OpenAI paired hosted agent infrastructure with a later option for local, customizable weights; and Zhipu strengthened a China-centered ecosystem of models and autonomous assistants.
For a production workflow, start with the least autonomous method that solves the problem. Use conventional automation or a stable API when possible, a browser agent for carefully bounded interface work, hosted models when operational simplicity dominates, and open weights only when control or customization pays for the added infrastructure. Treat a free assistant as an experiment until its geography, data terms and enterprise controls are verified.
The lasting significance of the 2025 announcements was therefore architectural, not merely promotional: AI competition was spreading across the browser, the developer platform, the cloud, the model-licensing layer and the consumer market at the same time.
Quick 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.
