The low-code and no-code space has been moving fast for the past several years, but 2025 was the year the ground shifted underneath it. AI stopped being a feature bolted onto existing platforms and started restructuring the entire category. New tools emerged that didn’t exist 18 months ago. A viral open-source project gave a glimpse of what truly autonomous AI assistants look like in practice. And the established players — Microsoft, Salesforce, ServiceNow — began racing to embed agents into everything they ship.
This article takes stock of where things stand in early 2026: what’s new, what’s genuinely useful, what’s hype, and how to think about it all.
The market in numbers Link to heading
The low-code/no-code market reached approximately $45.5 billion globally in 2025, growing at a 28% compound annual rate since 2020. Gartner forecasts the market to exceed $30 billion for the development platform segment alone by end of 2026. More striking: Gartner predicts that 40% of enterprise applications will feature task-specific AI agents by end of 2026, up from less than 5% in 2025.
70% of new applications are expected to include low-code or no-code components by 2026. And 75% of large enterprises will have adopted at least four low-code tools. These aren’t fringe tools anymore — they’re mainstream infrastructure.
The challenge, as always, is separating the platforms actually delivering value from the ones selling a vision of the future that doesn’t quite work yet.
The biggest shift: from automation to agentic systems Link to heading
The defining trend of 2025-2026 is the move from automation (if X happens, do Y) to agentic AI (achieve this goal, figure out the steps yourself).
Traditional workflow automation is deterministic. You map out every branch, every condition, every possible input. The system executes exactly what you configured. This works well for stable, well-defined processes — invoice approval, onboarding checklists, scheduled reports.
Agentic systems work differently. You give the agent a goal — “handle tier-1 support tickets” or “keep our inventory database up to date” — and the agent figures out how to achieve it, using whatever tools it has access to. It can take multiple steps, recover from failures, and adapt when something unexpected happens.
According to Deloitte’s 2025 Emerging Technology Trends study, 30% of organizations are currently exploring agentic options and 38% are piloting them — but only 11% have them actively running in production. The gap between pilot and production remains real. Governance, auditability, and the question of what the agent is allowed to do without asking permission are the main obstacles.
The platforms shaping the category in 2026 Link to heading
Microsoft Power Platform — more agent, less form Link to heading
Microsoft has gone all-in on agents. Copilot Studio, which I covered in a previous article, has become the central hub for building and deploying organizational AI agents. The 2025 updates added generative orchestration (the agent decides its own action sequence), MCP server support for connecting to custom APIs, and multi-agent coordination where specialized agents hand off tasks between each other.
Power Automate now has an “agentic flows” mode alongside traditional cloud flows. The distinction: traditional flows follow a fixed path you designed; agentic flows are given an objective and a set of available actions, and the AI determines the execution sequence at runtime.
The Power Platform still has its known limitations — licensing costs, Dataverse dependency for the most capable features, the ceiling on canvas app complexity. But as a platform for building agents that work inside the Microsoft 365 ecosystem, it’s currently unmatched in its integration depth.
Make and Zapier — playing catch-up with AI Link to heading
Both Make and Zapier have added AI capabilities, but neither has fundamentally reinvented their core product. Make’s AI steps allow you to incorporate LLM calls inside scenarios — useful, but you’re still defining the flow structure manually. Zapier’s “AI actions” follow the same pattern.
What both platforms do still offer is the widest app connectivity in the category. If you need to connect 15 different SaaS tools without writing any code, Make and Zapier remain the most pragmatic options. For straightforward automation that doesn’t require agentic reasoning, they’re fast to implement and well-understood.
The risk for both: as AI coding tools lower the bar for custom integrations, and as n8n’s self-hosted model gains traction, the per-task pricing model starts looking increasingly expensive for teams at scale.
n8n — the developer favorite keeps growing Link to heading
n8n passed 90,000 GitHub stars in 2025 and added native LLM node support, making it the most capable self-hosted option for teams that want both automation and AI in one place. The ability to chain AI calls, transform data with JavaScript, and run everything on your own infrastructure — with no per-task costs — is a compelling combination for technical teams.
The limitation remains: it requires someone who can operate infrastructure. For organizations without a technical owner, n8n is a maintenance burden waiting to happen.
Retool, Bubble, Webflow — specialized tools finding their lanes Link to heading
Retool remains the strongest option for internal data tools connected to databases and APIs — think admin panels, operations dashboards, data editors. Its AI-assisted component generation has sped up the building process noticeably.
Bubble continues to serve the no-code startup segment: full web applications with databases, user authentication, and complex logic — without writing code. The performance ceiling has improved, though it still struggles with truly high-traffic applications.
Webflow has cemented itself as the standard for marketing sites and content-driven applications where visual design quality matters. Its logic and CMS capabilities have expanded, though it remains fundamentally a design and publishing tool rather than an application builder.
ClawdBot: what a truly autonomous personal assistant actually looks like Link to heading
In late 2025, an open-source project by Austrian developer Peter Steinberger went viral on GitHub. Originally called ClawdBot, then Moltbot after a trademark challenge from Anthropic, and now OpenClaw — the project gave many people their first real glimpse of an AI assistant that actually acts rather than just responds.
The concept: a self-hosted AI agent that lives in your existing messaging apps. You send a message on WhatsApp, Telegram, Slack, Teams, or iMessage, and it doesn’t just answer — it executes. It can manage your calendar, draft and send emails, check you in for flights, run terminal commands on your computer, control smart home devices, and invoke custom scripts. It has persistent memory across sessions and can proactively reach out to you — a morning briefing, a reminder, a task status update.
The project accumulated over 145,000 GitHub stars, which is an extraordinary adoption signal. The underlying model recommendation is Claude Opus — the agent’s reasoning quality depends heavily on the model powering it.
What makes it relevant for businesses Link to heading
For a technically capable individual or team, OpenClaw represents something qualitatively new: an AI assistant with actual computer access that operates continuously, not just when you open a chat window. The business scenarios this unlocks:
- Executive or team assistant replacement for routine tasks. Meeting scheduling, email triage, status report compilation, data lookups across internal systems — delegated to the agent by message, executed in the background.
- Personal automation hub. Instead of building a Power Automate flow every time you want to automate something, you describe what you want to the agent and it figures out how to do it using the tools you’ve connected to it.
- Cross-system coordination. The agent can interact with multiple systems — your CRM, your calendar, your inbox, your project tracker — in a single task, without you having to build an explicit integration between each pair.
The serious caveats Link to heading
The OpenClaw vision comes with real risks that are worth stating plainly.
Security is the primary concern. The agent requires access to your system to execute actions — in many configurations, this means it can run terminal commands, access files, and interact with sensitive data. On a shared or corporate network, the security implications of running an autonomous agent with system access need careful evaluation. This is not a tool to hand to everyone in an organization and walk away.
Cost and complexity. Running OpenClaw properly requires hardware or a VPS, plus API costs for the underlying model. Estimates run $25–125/month depending on usage and infrastructure choices. It also requires Node.js knowledge, Docker proficiency, or Linux administration — the setup is not consumer-grade.
Reliability and the cost of mistakes. An agent that can send emails and execute commands can also send the wrong email and execute the wrong command. Without careful configuration of what the agent is and isn’t permitted to do autonomously, the blast radius of a misunderstood instruction can be significant.
Not ready for non-technical users. The current state of OpenClaw is firmly in developer-and-power-user territory. The configuration, maintenance, and incident response requirements are real. An enterprise deployment would need significant wrapping in policy, access controls, and oversight tooling.
The emerging framework: governed agentic automation Link to heading
The pattern that serious enterprise deployments are converging on in 2026 is what analysts are calling governed agentic automation — AI agents with defined scopes, explicit permission boundaries, and human escalation paths built in from the start.
The key principles:
- Define what the agent can and cannot do autonomously. Agents that can read and summarize data are lower risk than agents that can write to production systems. Most organizations are starting in “read and recommend” mode before graduating to “act on approval” and eventually “act autonomously within defined limits.”
- Audit trails on everything. Every action an agent takes should be logged in a way that a human can review. This is both a governance requirement and a debugging tool when things go wrong.
- Human escalation by design. When the agent encounters something outside its configured scope, the correct behavior is to pause and ask — not to guess. The failure mode of an overly conservative agent (it asks too often) is much better than the failure mode of an overly autonomous one.
- Start with low-blast-radius tasks. Internal knowledge retrieval, report generation, meeting summaries — these are good starting points because the cost of an error is low. Customer-facing or financial actions come later, if at all.
What this means practically Link to heading
If you’re deciding what to invest in right now, the landscape in early 2026 looks like this:
For internal process automation in Microsoft environments: Power Platform and Copilot Studio are the clear choice. The agent capabilities are maturing fast, the integration with M365 is unmatched, and the governance tooling is enterprise-grade.
For cross-platform automation without Microsoft lock-in: Make remains the best balance of visual interface, data transformation capability, and price. n8n is better if you have the technical capacity to self-host.
For personal productivity and individual-level AI assistance: Tools like OpenClaw represent where things are heading, but they require technical investment to run safely and reliably. For most users, the managed alternatives — Microsoft 365 Copilot, Notion AI, dedicated AI assistants — remain more practical despite offering less autonomy.
For building internal tools and dashboards: The combination of a developer with AI coding tools (Cursor, Claude, GitHub Copilot) and a modern tech stack is increasingly competitive with Retool for teams that have even one capable developer. For truly non-technical builders, Retool still wins.
The honest summary: the tools are ahead of most organizations’ ability to deploy them safely and get consistent value from them. The bottleneck in 2026 is no longer “do the tools exist?” It’s “do we have the governance, the processes, and the people to use them responsibly?”
That gap is where the real work is.
Sources: Gartner on AI Agents in Enterprise Apps · Deloitte Agentic AI Strategy · OpenClaw on GitHub · Low-Code/No-Code Trends 2026 – Codewave