QX Labs vs Tasklet: Two Different Bets on AI Agents

Tasklet (tasklet.ai) builds cloud agents that re-reason from scratch on every run: you describe the job in chat, and the agent plans its own path each time, inside a sandbox with code execution and a browser. QX Labs treats that as one tool among several. You get conversational Agents, but also Flows that lock the repeatable steps down deterministically, Grids that apply one piece of work to thousands of records in parallel, and Knowledge Vaults that turn your internal documents into an AI Brain your agents draw on, with citations. Tasklet suits an individual or small team who wants a clever agent running odd jobs in the cloud. QX suits teams who need those agents to behave the same way on run four hundred as on run four, at volume, grounded in company data.

We build QX, so weigh this piece accordingly. Everything we say about Tasklet below comes from their public site, docs, and press coverage, and we link the sources so you can check them.

TL;DR: the five differences that matter

• Where the work gets delegated. Tasklet lives in its web app; agents fire on schedules, events, and webhooks. QX agents also take work by @mention in Slack, Microsoft Teams, WhatsApp, and email, which is where most teams actually hand things off.
• Re-planning vs guardrails. Tasklet's design principle is that the workflow disappears and the agent reasons about what to do on each run. QX Agents work similarly, whereas Flows lets you construct workflows with tighter guardrils: deterministic nodes where behaviour must be identical every time, agentic nodes where judgement helps.
• Single runs vs volume. Tasklet executes an agent per trigger. QX offers trigger-based runs too, but Grids exist for the jobs Tasklet has no primitive for: enriching 5,000 leads, reading 800 contracts, scoring an entire market in one pass.
• Attached context vs an AI Brain. Tasklet offers knowledge bases you attach to agents. QX Knowledge Vaults index your org's documents, stay synced as they change, and return cited answers inside any agent, flow, or grid.
• Cost predictability. Reasoning on every run is token-hungry, and Tasklet's credit burn scales with the "intelligence level" a task needs. QX shows estimated credit costs before you run anything at scale, and deterministic steps don't pay a reasoning tax.

Where does Tasklet come from?

Tasklet's pedigree is worth knowing. It was founded by Andrew Lee, who co-founded Firebase (acquired by Google) and the AI email client Shortwave, alongside early Firebase engineer Jonny Dimond. The product came out of beta in October 2025, and the company has reported fast early growth and a $20M raise since. The team is small, senior, and moves quickly.

The product reflects a strong thesis, which Lee has summed up as betting on the models: instead of wrapping AI inside a fixed workflow, let the workflow go away and have the agent reason about what to do. In practice that means:

• You build by chatting. Describe the job, and Tasklet creates the agent. No canvas, no flowchart.
• Every agent gets a cloud sandbox. Each one runs 24/7 server-side with code execution, file handling, and a browser it can drive when no API exists.
• Connections are broad. Thousands of app integrations, plus any MCP server, plus automatic connection building against arbitrary HTTP APIs.
• It can build small apps. Ask for a tracker or dashboard and Tasklet generates the software as part of the job.
• Teams get the basics. Shared agents, central credential management, per-member usage limits, and SOC 2 compliance.

For a certain kind of job, this is a great shape: a monitoring agent, an inbox triage agent, a nightly report, a scraper for a portal with no API. Describe it once, let it run, glance at its reasoning when it surprises you.

What does QX Labs do differently?

QX starts from the same observation (hand-wired workflow builders are brittle and slow to build) but lands on a different conclusion. An agent that re-derives its plan on every run is the right tool for judgement-heavy work and the wrong tool for a business process that must not drift. So QX gives you both, in one workspace, and lets them share everything: tools, data, and each other.

A QX Agent is briefed in plain English like a Tasklet agent, connects to 1,000+ apps, and remembers what it learns about your organisation between runs. But it's one of four primitives rather than the whole product:

• A Flow chains steps where each node is either deterministic (call this API, write this row, send this email if the score clears 7) or agentic (classify, summarise, decide). Approval gates sit in front of sensitive actions. The repeatable spine of the process is fixed; the judgement calls are delegated.
• A Grid is a table where every column does work: web research, a CRM lookup, a scoring rubric, a custom agent, a query against your knowledge. Configure the column once, run it down thousands of rows in parallel, and inspect any cell to see how its result was produced.
• A Knowledge Vault is the AI Brain underneath all of it: your internal documents indexed, mapped, and kept current as sources change, so the right context is pulled into whatever is running, and answers arrive with citations back to the source.

The pieces interlock. A flow can call an agent, an agent can sit in a grid column, and everything can query the same knowledge. And because agents, flows, grids, and skills can all be published publicly, QX builders rarely start cold: there's a community of shared examples to adapt instead of a blank chat box.

QX Labs vs Tasklet side by side

	Tasklet	QX Labs
Core idea	The workflow goes away; the agent re-plans each run	Judgement where it helps, determinism where it matters
Built by	Chat description → cloud agent	Plain-English description → agent, flow, or grid, refined in conversation
Repeatable processes	Agent reasoning + triggers (schedule, event, webhook)	Flows with deterministic nodes, branches, and approval gates
Work across many records	One agent execution per trigger	Grids: thousands of rows processed in parallel
Company knowledge	Knowledge bases attached for context	Knowledge Vaults: continuously synced AI Brain, cited answers
No-API tools	Sandbox browser / computer use per agent	1,000+ integrations incl. long-tail research and enrichment APIs
Models	Routed by "intelligence level" (Basic → Genius), one subscription	You pick the model per task; bring your own keys
Delegation surfaces	Web app, triggers, webhooks	Slack, Teams, WhatsApp, email, web app, API
Sharing	Shared agents within a team	Agents, flows, grids, and skills publishable publicly
Entry pricing	Free tier (daily bonus credits, capped triggers); paid from $25/mo	Free plan, every feature included; credits pooled workspace-wide, no per-seat charge

Tasklet details from tasklet.ai and press coverage at the time of writing; both products ship weekly, so confirm current specifics on each pricing page.

Should an agent re-plan every run?

This is the real fork in the road between these two products, and it's worth thinking through rather than treating as a feature checkbox.

Tasklet's answer is yes. The upside is real: an agent that reasons freshly can absorb messy inputs, recover from failures, and handle cases nobody anticipated when the automation was written. Reviewers have praised exactly this flexibility, and for exploratory or low-stakes work it's the right default.

The costs show up in production. Fresh reasoning means each run can take a different path, so an automation that touches customers, money, or your CRM behaves probabilistically rather than predictably. It's slower than calling an API directly. And you pay for the thinking every single time, which is why token-intensive runs and less predictable bills are the most common caveats in Tasklet coverage.

QX's answer is: it depends on the step. Invoice matching should never be creative. Deciding whether an inbound lead deserves a same-day call should be. A QX Flow encodes that distinction explicitly, so the deterministic 80% of a process runs fast, cheap, and identically every time, while the agentic 20% gets model judgement with a human approval gate wherever the blast radius is large. Every run leaves a trace you can open step by step.

What happens when the work is 5,000 rows long?

Ask each product to research one company and both do fine. Ask for the same brief across an entire market and the difference is structural. Tasklet's unit of execution is the single agent run; its free tier even caps executions per trigger. There's no native construct for "do this identical piece of work to every record in this list and let me compare the outputs."

That construct is exactly what a QX Grid is. Rows are your records. Each column is a worker. One person points a scoring column at 5,000 leads, validates the first fifty, checks the estimated credit cost, and then lets the rest run in parallel. Outputs are consistent because every row got the same logic, and any cell can be audited afterwards. Teams that live on research, enrichment, or document extraction spend most of their QX time here, and it's the piece a single-agent architecture can't emulate without spawning thousands of independent runs.

How does each platform know your business?

Tasklet lets you attach knowledge bases to agents for context, which covers the basics: an agent that needs your pricing sheet or tone-of-voice guide can have it.

QX treats company knowledge as infrastructure rather than an attachment. Knowledge Vaults build an AI Brain from the documents across your org: they index the content, keep the index synced as source files change, and map it so relevant material is pulled into context automatically, whichever surface the question comes from. A grid column can interrogate your past deal memos row by row. A flow can check policy mid-pipeline. An agent answering in Slack cites the exact document it drew from, so a sceptical colleague can click through and verify. If your evaluation includes the question "how will this thing use what our company already knows?", this is the section to test hardest in both products.

When is Tasklet the right choice?

Plenty of buyers should pick it. If you're an individual or a small team automating personal-productivity-shaped work (email triage, meeting prep, monitoring, scheduled reports), Tasklet's chat-first build and always-on sandboxes get you moving in minutes for $25 a month. If your workflow depends on a tool with no API, its per-agent browser automation handles what integration lists can't. If you want an agent to spin up a quick internal tracker or dashboard as a by-product of the work, app generation is a distinctive touch. And the founding team's Firebase heritage shows in the developer-friendly details, like first-class MCP support and automatic connections to arbitrary HTTP APIs.

The trade-offs to go in with open eyes about: costs that scale with how hard the agent has to think, run-to-run variability on processes you'd rather were boring, and no bulk primitive when the job is a list of thousands.

When is QX Labs the right choice?

Choose QX when agents stop being a personal productivity trick and start being how the team operates. That's when the requirements change shape: the same logic applied to every record, pipelines that behave identically while unattended, approval gates before anything irreversible, answers that cite the company documents they came from, and credits pooled across the workspace so adding a colleague costs nothing. It's also the better home if you want to start from proven, publicly shared agents, flows, grids, and skills rather than describing everything from scratch, or if procurement wants model choice with your own API keys, role-based access, audit logs, and the option of an isolated tenant.

A fair one-line summary: Tasklet is a bet that the model's reasoning can replace the workflow. QX is a bet that teams need both, connected to the same tools and the same brain.

Try the comparison on your own work

Reading only gets you so far with either product. Pick the process you most want off your plate, then book a demo and we'll build it live in QX, or start free and build it yourself; the free plan includes Agents, Grids, Flows, and Knowledge Vaults. If you're still mapping the market, our guide to the best AI agent platforms and comparisons with Lindy and Twin cover the neighbouring options.