Langdock Review From Daily Use: What One Month of Rollout Changed for Us
Von Manuel Gick·Certified AI & Automation Expert
Langdock is the AI platform on which a team runs chat, knowledge bases and its own agents, GDPR-compliant and hosted in the EU. This Langdock review comes not from a trial account but from our own company: all of our 40-plus own AI agents run in Langdock, alongside further agents and AI workflows elsewhere in the business.
Most Langdock reviews are feature lists. This one is about what a rollout actually looked like in a company of around 20 people: a pilot phase with three developers, a kit for the rest of the team, a few rules we gave ourselves after making mistakes, and the numbers from one month that tell us whether the tool has arrived in everyday work. The short version first: in August the chat still had the largest share of our usage, at 43 percent. In September seven out of ten messages ran through agents.
What did it look like before the rollout?
Langdock had been in use for a while, as the team chat and for the first agents, among them the bookkeeping agent that answers questions about receipts and ledger entries straight from our accounting system, bank account and mailbox and, when asked, posts entries itself. How we automated our own bookkeeping to around 90 percent is in Automate Your Bookkeeping. What was missing was depth. The platform was used, but mostly the way you use a chat tool: type a question, read the answer, type the context again next time.
Langdock's usage analytics for August (3 August to 4 September) show it clearly. Of around 3,500 messages in the month, 43 percent ran through the chat and 38 percent through agents, the rest through projects. 38 different agents were in use, and twelve of the 13 users were active. So the lever was not more users but the way the platform was used: a large part of the work still ran as question and answer in the chat rather than through agents that complete tasks with the connected tools. The licence was there, the tool worked, and still the biggest lever stayed untouched. Our impression is that many teams get stuck at exactly this point.
Why we stayed with Langdock regardless has less to do with the chat than with everything else: a company account instead of private ones, roles and permissions, a data processing agreement, ISO 27001, hosting in the EU, and the agent feature that makes a helper built once available to the whole team. Transparency: the Langdock link above is a referral (affiliate) link. If you use it, we may receive a commission. The price stays the same for you. What the platform fundamentally is and who it is built for is covered in What is Langdock?.
How did we structure the rollout?
We planned about six weeks in three stages: pilot phase, preparation, rollout for everyone. A colleague led the pilot phase and the preparation, and will now focus mainly on Langdock rollouts for clients.
Stage 1: three pilots, three personal assistant agents
For the pilot phase we picked three developers with different starting points: one used Langdock almost only as a chat, one already worked mostly with agents, one sat in between. They did not get a finished solution, they got material: which integrations exist, which skills an agent can be given, what projects and templates are for. The rest they built themselves.
All three built a personal assistant agent first, without anyone prescribing it: an agent that knows their own work, is connected to their own tools and hands tasks on to specialised agents. Alongside it, each pilot built something of their own: an agent that compares yesterday's work with today's state and updates the documentation in the project tools, an agent that breaks a client request from Slack or email down into possible meanings and proposes the safest starting point, and first building blocks for an agent builder that is meant to make delegation between agents smoother.
After four weeks we compared the three pilots' usage before and after. For the chat user, agent usage rose by around 60 percent. The second doubled his total messages, because a daily routine now feeds his assistant the tasks of the day every morning. The third ended up working almost entirely through agents. That was the signal we needed: the effect did not depend on one person, it depended on the pattern.
Stage 2: turning the pattern into a kit
The honest lesson from the pilot phase: each pilot's assistant was tailored closely to its owner, and customising at that depth can eat a whole working day. You cannot ask that of colleagues whose main job is delivering client projects. So we shortened the path without shortening the result:
- A template for the personal assistant agent. Everyone duplicates it, connects their own accounts and answers a few questions in Langdock's build function: what should the assistant handle, what needs confirmation, where does it delegate. The two pilots who tested it needed ten to fifteen minutes. Anyone who wants to can customise as deeply as they like afterwards.
- An overview of all skills and agents, sorted by task, so nobody has to click through the list in the platform to know what already exists.
- Suggested sub-agents the assistant delegates to: the agent that writes user stories, the knowledge agent on our database, the specialist for our Make.com scenarios.
- A rule for model choice. In a series of tests, the colleague leading the rollout found that the platform's automatic model selection gives very good results for routine tasks at a fraction of the usage of the large model, and shared this with the pilots. Since then the default is the automatic setting, with the large model reserved for tasks that genuinely need the reasoning, such as building a complete interface.
- One AI champion per development team who answers questions in the group, so the rollout does not depend on one person.
Stage 3: two weeks of rollout, not two months
We deliberately set only two weeks for the rollout: one week to set up your own assistant, one week to bring it into your daily work. The reason is practical: for everyone the rollout is a side task. Client projects and bug fixes come first, and a rollout that takes longer than two weeks loses against everyday work. It started at the end of September. What the same rollout covers for clients is on our Langdock page.
The rules we gave ourselves, some of them after mistakes
Some rules were in place before the rollout, others came from mistakes:
- Everyone builds their own agents; shared agents get reviewed first. Personal helpers belong to the person who uses them. Only when an agent is released for others does someone take a look.
- Agents may read on their own; changing and sending gets confirmed by a person. In Langdock you can define per integration which actions need approval. For us: change and delete always, read never.
- Emails only as drafts. At first, agents were allowed to send mails directly. There were repeated mistakes and odd phrasings. Since then they write drafts, a person reads and sends.
- Own credentials instead of shared keys. Everyone connects their tools with their own access. That way an agent only has the rights of its person, and an access can be revoked individually without breaking other agents.
How to set up roles and permissions before the first agent touches real data, and the general path from the first login to the first agent, is covered in Langdock Rollout: The Practical Guide.
What changed in one month?
The numbers come from Langdock's usage analytics for our team, 13 users, September (7 September to 4 October) compared with August (3 August to 4 September). What gets measured is messages in the platform, not hours saved. That matters: these numbers show whether a tool has arrived in everyday work, not what it delivered. September covers the pilot phase, the preparation of the kit and the first week of the team rollout, so these numbers show the effect of the whole programme so far, not of the two rollout weeks alone.
| Metric | August | September | Change |
|---|---|---|---|
| Messages in total | 3,502 | 5,276 | +51% |
| Share of agents in all messages | 38% | 71% | +33 points |
| Share of chat | 43% | 13% | −30 points |
| Messages per day | 106 | 188 | +78% |
| Messages per active user | 292 | 440 | +51% |
| Different agents in use | 38 | 72 | +89% |
| Tool actions by agents (mailbox, files, Slack, calendar, database) | 2,554 | 6,093 | +139% |
| Users above 500 messages | 0 | 4 | +4 |
Two things matter more to us than the percentages. First, the shift: agents went from fewer than four in ten messages to seven in ten. Chat fell from the biggest share to the smallest. For us that is the clearest sign that Langdock no longer serves only as a search box but as a tool we hand tasks to. Second, the breadth: twelve of the 13 users were active in both months, eight of them increased their usage, and the number of agents actually used almost doubled. Not only the heavy users grew, the middle did too: the median user went from 331 to 403 messages.
To be fair, there is another side too. The share of the three heaviest users in total usage rose from 36 to 48 percent. Part of the growth therefore comes from a few people, and four users used the platform less than in August. For their tasks we have evidently not yet found a process where an agent takes noticeable work off their hands. Finding it is the next job of the rollout, which is exactly what the template and the champions are for. And September, at 28 days, is shorter than the measured August at 33, which is why messages per day are a fairer figure than the totals.
What we take from one month of using Langdock
- The right metric is the agent share, not the number of licences. A team that sends 70 percent of its messages through agents is handing tasks over to them. A team with many chat messages is mostly using a search box.
- The personal assistant is the entry point that sticks. None of the pilots had to be pushed: as soon as they saw a colleague's assistant, they wanted their own, and one colleague outside the pilot group asked for access on his own initiative. A template you can set up in fifteen to twenty minutes turns that into a rollout instead of a hobby.
- Rules before reach. Draft instead of send, approval for changes, own credentials. These three rules take hardly any time in daily work and keep the mistakes we made early on from happening again.
- The automatic model setting is enough for most tasks. Give every agent the largest model and you pay many times the usage for work the automatic setting handles well.
- Plan two weeks, not two months. Not because it is no work, but because for everyone the rollout is a side task, and a rollout that drags on loses against everyday work.
More on the costs that come on top of the licence, from setup to ongoing support, is in Langdock Pricing. Which agents typically come first in small and mid-sized companies is covered in AI Agents for SMEs.
What happens next for us?
The rollout is not finished, it is running. The next job is breadth: bringing the colleagues who have used the platform little so far to where the pilots are today, with the template and a champion at their side. We keep measuring with the same two numbers we plan with for client rollouts: active users per week and the share of usage that runs through agents rather than the chat.
Our Langdock rollout for clients follows the same logic: a kick-off sprint in which roles, permissions, the first process and the first agents are set up, followed by support over several months, because the real work starts after the kick-off. We tested the pattern in this article, pilots, template, champions, rules, measurement, on ourselves first. It is the basis on which we plan client rollouts.
Frequently Asked Questions
How long does the Langdock rollout take at your company?
We planned about six weeks in three stages: a pilot phase with three developers, preparing the kit and a two-week rollout for the whole team, which started at the end of September and is still running. The frame is deliberately short because the team already used the platform as a chat, the template cuts the setup to fifteen to twenty minutes, and a rollout that drags on loses against client work.
What is a personal assistant agent, and why does everyone on your team have their own?
An agent that knows one person's work, is connected to their tools such as mailbox, calendar, Slack and project tools, and hands tasks on to specialised agents. Everyone has their own because the integrations run through personal access and the tasks differ by role. A shared template means the setup still takes only around fifteen to twenty minutes.
Which rules apply to agents in your company?
Four. Everyone builds their own agents, shared agents get reviewed first. Agents may read on their own, changing and sending gets confirmed by a person. Emails are created only as drafts. And everyone connects their tools with their own access, not with a shared key. One of them, emails only as drafts, came from our own mistakes; the others come from everyday practice and from a security check before the rollout.
What did Langdock cost you?
The licence is priced per seat per month; current prices and a worked example are in Langdock Pricing. Next to the licence there is the work around it: the pilot phase, the kit, the champions and the time everyone puts into their first agent. That is the work we take on for clients in a Langdock rollout. If you budget only for the licence, that work is missing from the plan.
Does Langdock replace Make.com or n8n for you?
No. We still build processes that run in the background without a person on automation platforms such as Make.com or n8n. Langdock is where chat, knowledge bases and the agents people work with live. The two interlock: an agent in Langdock can analyse and adjust our Make.com scenarios, and the scenarios handle the steps that follow.
How do you know a rollout is working?
From two numbers Langdock provides in its usage analytics: how many people actively work with it per week, and how large the share of usage through agents is compared with the chat. If the first number rises and the second shifts towards agents, the tool has arrived in everyday work. If both stand still, more training will not help; a concrete process that eats time will.
The next step
If you are weighing up whether an AI platform in your team will get beyond the search box, the honest first question is not about the tool but about the process that eats the most time in your company. That is exactly what we find in the free Time Potential Analysis: 45 minutes, no sales pitch, and you leave with your three biggest time sinks and a recommendation on whether Langdock, an automation or something else entirely is the right first step.
