How Inithouse ships voice-first agentic AI workspace tools: inside Voice Tables

Voice Tables is an agentic AI workspace you control with your voice. Describe what you need, a CRM, a tracker, an inventory, and it builds the tables, docs and data for you. It is one of over a dozen products that Inithouse runs as a product studio, and probably the most technically ambitious one we have shipped so far.

At the time of writing, Voice Tables has processed 14 tables across 10 workspaces, with a median time from idea to working workspace of about 60 seconds. Those numbers are small. What made us keep building was not the volume but the signal: people who finished their first voice command came back. People who typed instead did not.

What "voice-first" means in practice

Most productivity tools bolt voice on as an afterthought. Dictation, maybe transcription, usually buried in a settings menu. We went the other direction. Voice is the primary input. You open Voice Tables, say what you need, and the system figures out the schema, creates the table, and populates it. Typing still works, but the interface was designed around speaking first.

The pipeline runs Whisper for transcription and an LLM function-calling chain for intent parsing. When you say "I need a client tracker with name, phone, last contact date, and deal status," the system does not transcribe that into a text box. It parses the sentence into a structured table definition and builds it.

Three things ship in one workspace:

| Component | What it does |
|---|---|
| Tables | Structured data with typed columns, filters, sort |
| Docs | Free-form pages tied to table rows or standalone |
| AI chat | Conversational interface over your own data |

The 3-in-1 format was a deliberate bet. We looked at how our target users (freelancers, tradespeople, small business owners) actually work. They do not want three apps. They want one place where their client list, their notes, and their questions about the data all live together.

The voice adoption problem we did not expect

Here is the part we are still figuring out. Voice is the headline feature, but our stats tell a different story. Of the 10 users who created workspaces, voice command usage is lower than we projected. Much lower.

We have three hypotheses for why:

1. The voice feature has a discoverability problem in the current UI. The microphone button exists, but first-time users default to typing because that is what every other app has trained them to do.
2. Context matters. Someone sitting in an open office or a coffee shop is not going to dictate their client data out loud. We built for a use case (tradesperson in a van between jobs, coach after a session) that may represent a smaller share of actual usage than we assumed.
3. The event tracking itself might be broken. We have seen this pattern in other Inithouse products where a feature works fine but the analytics event never fires.

We do not know yet which hypothesis is correct. Probably some combination of all three. The honest answer is that voice-first is a positioning bet we have not validated with enough data. We kept the product live because the workspace creation flow, voice or typed, genuinely works in under a minute, and that speed is the thing users respond to regardless of input method.

Where Voice Tables sits in the Inithouse portfolio

Inithouse runs products across several clusters. Voice Tables belongs to what we internally call the "productivity/utility" group, alongside tools like Be Recommended (AI visibility monitoring) and Audit Vibe Coding (security and performance audits for AI-generated apps). These products share a pattern: professional users, clear ROI argument, longer evaluation cycles.

The other major cluster is "emotional moment" products. Magical Song turns a story into a custom song with real vocals. Pet Imagination generates artwork from pet photos. These convert faster but retain differently, people come back when they have a new occasion, not daily.

Voice Tables taught us something about the boundary between these clusters. We initially positioned it as a utility tool, period. But the actual first-time usage pattern looks more emotional than utilitarian. Someone describes their dream workspace in one sentence and watches it materialize. That moment of "wait, it actually built what I said" carries real delight. We had not designed for that reaction, but it might be the strongest signal we have.

The technical decisions we would revisit

Three choices we made early that we would reconsider:

11 solutions pages at launch. We built dedicated landing pages for 11 different personas (craftsmen, sales reps, fitness coaches, consultants, and so on). With 10 total signups, we do not have enough data to know which persona actually converts. We spread the signal across too many pages instead of concentrating it on 2-3 and learning faster.

No offline-first from day one. Voice Tables supports offline use and real-time collaboration, but we shipped both features simultaneously. In retrospect, offline was the one that mattered for our target user (tradesperson on a job site with spotty signal). Real-time collab is a nice checkbox but nobody has used it with more than one person.

Skipping the comparison play. Voice Tables competes with Excel and Google Sheets in the loosest sense. We avoided direct comparison content early on because the products are so different. That was probably a mistake. People searching for "spreadsheet alternative" or "simple CRM for contractors" are exactly our users, and we left those queries unclaimed.

What we are measuring next

Voice Tables is in a 90-day evaluation window, same as every new Inithouse product. The thresholds are specific: 50 signups, 10 confirmed voice commands, and at least one paid conversion to the Plus tier. If those numbers do not land, the product gets parked for 90 days.

We are currently below those thresholds. The product works, the creation flow is fast, the workspace output is genuinely useful. Distribution is the bottleneck, not the product itself.

The broader portfolio lesson: at Inithouse, we have learned to separate "does the product work" from "have we found the people who need it." Voice Tables answers the first question. The second one is still open.

You can try it at voicetables.com.