Your SaaS Dashboard Is Becoming a Spectator

In February 2026, $2 trillion was wiped from software stocks in under a month. Atlassian dropped 35%. Salesforce fell 28%. The trigger? Anthropic launched plugins for Claude Cowork, and Wall Street suddenly realized that AI agents can do things that used to require logging into a dashboard, navigating a sidebar, clicking through three modals, and hitting submit.

The selloff was dramatic, probably overdone, and the market will recover. But the underlying signal is real: the way humans interact with software is changing faster than at any point since the smartphone.

This is not a post about AI replacing software. It's about AI replacing the interface to software. That distinction matters — and it's the reason we're building Expensicat the way we are.

The dashboard had a good run

For twenty years, every SaaS product followed the same playbook: put data in a database, build a web interface on top of it, charge per seat per month. The interface was the product. You paid for Salesforce not because of the PostgreSQL instance underneath, but because of the dashboards, the pipeline views, the reports you could generate by clicking the right combination of filters.

That model rewarded complexity. More features meant more menus. More menus meant more training. More training meant stickier customers. The SaaS feedback loop was: make the dashboard indispensable, then charge for access to it.

AI breaks this loop.

The evidence is already here. Ramp reports 99% of expense reports processed without a human opening the app. Brex claims similar numbers. These are enterprise tools built for companies with hundreds of employees and dedicated finance teams.

But there's a gap in this story. All these companies are building for the enterprise — 50+ seats, dedicated finance departments, corporate card programs. If you're a freelancer in Lisbon, a two-person agency in Berlin, or a small business owner in Lagos, Ramp's AI agent isn't solving your problem. You don't have an expense policy to enforce. You have a shoebox of receipts and a quarterly tax deadline.

That's who we're building for.

From clicking to talking

The shift from GUI (graphical user interface) to CUI (conversational user interface) is happening everywhere. Expensify launched a "hybrid contextual AI expense agent". Salesforce built Agentforce. Intercom's Fin now resolves over 50% of customer queries without human involvement.

The pattern is consistent: the conversational layer does the work, the visual layer reviews the results. Software becomes something you talk to, not something you operate.

In Expensicat, this is already live. Our AI assistant sits inside your workspace. Ask it "what did I spend on software this quarter?" and it pulls the numbers. Tell it "categorize the last 30 transactions" and it does — with explanations for each decision you can review and correct. It learns from your corrections. Over time, it stops needing them.

The difference between this and a dashboard is fundamental. A dashboard requires you to know what you're looking for and where to find it. A conversation requires you to ask a question. One demands expertise with the tool. The other demands nothing.

The multi-interface future

Once you accept that the primary interface is no longer a web browser, the question becomes: what is the interface?

The answer is: whatever is closest to the user at the moment they need something done.

WhatsApp. In Brazil, a fintech called Magie has processed over R$1 billion (~$200 million) in transactions without a website or mobile app. Their entire product runs inside WhatsApp. With over 2 billion monthly active users globally, WhatsApp isn't a messaging app anymore. It's an operating system for commerce.

This is exactly why we're building Expensicat's WhatsApp integration. Picture this: you're at a client dinner, you snap a photo of the receipt, send it to Expensicat on WhatsApp. In seconds it's categorized, the VAT is extracted, and it's filed under the right project. No app to open. No login. No "I'll do it later" that turns into a lost receipt at tax time.

But it goes further than receipts. Ask "how much did I invoice this month?" and get the answer in the same WhatsApp thread. Get a notification when a payment lands or an invoice goes overdue. Your finance tool lives where your conversations already happen.

Slack and Teams. The workspace is already where decisions happen. SaaS that can surface inside those conversations — approving an expense, flagging an anomaly, answering a budget question — removes the context switch entirely.

Voice. A 2026 report found that 55% of consumers now use voice to interact with AI. Only 29% of companies have deployed voice AI. That gap is closing fast.

MCP servers. This one deserves its own section.

The point is this: "interface" used to mean "the screens we built." Now it means "any surface where a user or an AI agent can interact with our system." An API is an interface. A WhatsApp number is an interface. A voice command is an interface. An MCP server is an interface.

SaaS companies that only have a web dashboard have one interface. The ones that will win have five or six, and they all talk to the same backend.

MCP: the USB-C of AI

Anthropic introduced the Model Context Protocol in November 2024. By December 2025, it had been donated to the Agentic AI Foundation under the Linux Foundation, co-founded by Anthropic, Block, and OpenAI, with support from Google, Microsoft, AWS, and Cloudflare.

There are now over 10,000 active public MCP servers. The protocol has been adopted by ChatGPT, Cursor, Gemini, Microsoft Copilot, and Visual Studio Code.

The analogy that keeps coming up is USB-C: a single standard that lets any AI model interact with any tool. Before MCP, every integration was bespoke. MCP standardizes this. Cloudflare ran an MCP Demo Day with Stripe, Linear, Atlassian, PayPal, Sentry, and others — all building remote MCP servers on the same protocol.

Expensicat already ships an MCP server. Connect it to Claude, ChatGPT, or any MCP-compatible client, and your AI assistant can query your expenses, create invoices, pull financial metrics, and check bank balances — all through natural conversation. You never open a tab.

For a freelancer, this means your AI assistant can prepare your monthly expense report, flag anything unusual, and draft your VAT submission — pulling data directly from Expensicat without you touching a dashboard. For a small team, it means the founder's AI can ask "are we burning more than last quarter?" and get a sourced, accurate answer in seconds.

This is the "headless SaaS" concept: the value lives in the data and the logic, not in the interface. The interface is whatever the AI agent happens to be using.

Why human-in-the-loop isn't optional

The most instructive AI story of the past year isn't a success story. It's a correction.

In early 2024, Klarna announced their AI chatbot was doing the work of 700 full-time agents. It handled 2.3 million conversations in its first month, cut resolution times from 11 minutes to under 2 minutes, and was projected to drive $40 million in profit improvement.

Then customer satisfaction dropped. By mid-2025, Klarna reversed course and started rehiring human agents. CEO Sebastian Siemiatkowski admitted that "cost was a predominant evaluation factor" in organizing support, resulting in "lower quality."

This is the correct lesson, especially in finance. When someone's money is involved, "99% accuracy" isn't good enough for 100% of cases. The remaining 1% is where trust is built or destroyed.

The right model isn't "AI does everything" or "humans do everything." It's:

• AI handles the routine. Categorize this expense. Match this receipt. Extract the VAT. Flag the duplicate.
• Humans handle the exceptions. This receipt is blurry. This vendor doesn't match anything in the system. This charge looks wrong.
• The dashboard becomes the review layer. Not where you do the work, but where you approve it.

This is how Expensicat works. The AI categorizes your transactions, matches receipts to expenses, detects recurring charges, and flags anomalies. But every decision is reviewable. You see the AI's reasoning, and you can correct it with a click. The AI learns. Next time, it gets it right.

We'd rather be right 100% of the time with your help than wrong 1% of the time without it. When it's your money, that 1% matters.

The pricing reckoning

If AI agents do the work instead of humans, per-seat pricing makes no sense. You're charging for the number of humans who log in, but the humans aren't logging in anymore.

IDC predicts that by 2028, pure seat-based pricing will be obsolete, with 70% of software vendors refactoring their pricing around new value metrics. By 2030, at least 40% of enterprise SaaS spending will shift to usage-, agent-, or outcome-based pricing.

The Shopify signal is relevant here too. CEO Tobi Lutke told employees in April 2025 that AI usage is now a "fundamental expectation" and that managers must prove AI can't do a job before requesting new hires. Shopify grew at least 21% per year since 2022 while headcount dropped from 11,600 to 8,100.

When your customers are hiring fewer people, selling them per-seat software is swimming against the current. When your customers are freelancers and small businesses who never had a "seat" to begin with, the model was broken from the start.

The SaaS industry isn't dying. But the interface layer is being rebuilt from scratch. The companies that recognize this — that an MCP server is as important as a login page, that a WhatsApp number is as valid as a dashboard, that AI agents are users too — will define the next era.

We're building Expensicat for that era. A web app with an AI assistant. A WhatsApp integration for the moments when opening a laptop isn't the answer. An MCP server so your AI tools can work with your financial data directly. Three interfaces, one system, same source of truth.

The dashboard still matters. But it's no longer the only door. And for most of the things you do with your money every day — logging expenses, checking balances, filing receipts — it shouldn't have to be.