Ask Sidekicks: How Fixify’s AI tackles the messy middle of IT

The cold, hard reality of IT work? It’s messy. Not all issues are straightforward. Some are complex. That can stretch out resolution times and leave users waiting as analysts wrangle around for the right context. 

Yet there’s a persistent fantasy that humans are obsolete when it comes to IT help desk work and robots can handle it all. Get the right robot and your reputation as an IT leader is safe because — don’t worry — LLMs totally understand the human experience. They’ll follow your outdated knowledge base, enforce your processes, and deliver a flawless user experience.

Right. And the printer never jams. 

At Fixify, we know better because we live in the daily grind: thousands of tickets a week, knowledge changing by the hour, and both humans and AI struggling to keep pace. Relying on humans alone is cruel. Relying on AI alone is reckless.

That’s why we built Ask Sidekicks, our smart assistant — ready anytime to provide guidance, explanations, or next steps when our analysts need it most. Not to replace people, but to bring AI where it actually thrives: in the messy middle, where context and coordination matter most.

What does Fixify’s Ask Sidekicks feature do?

Sidekicks are our analysts’ AI teammates that live right where they work. We built Ask Sidekicks with the goal of always having the most relevant and up-to-date context available to it. That means, when an analyst has a question, the Sidekicks already have the context they need stitched together: machine IDs, triage reasons, ticket history, sentiment signals, automations, even task execution logs. Instead of analysts wasting time pulling data from five different places, Sidekicks surface it instantly.

And because the Sidekicks already have the context, questions don’t have to be perfectly worded prompts. Analysts can simply ask in natural language — “What should I do next?” or “Why did this fail?” — and our Sidekicks respond with a recommendation, backed by explanations and citations they can trust.

Behind the scenes, it’s not just one model pretending to know everything. You can think of each of our Sidekicks as a team of specialized agents. Sometimes one Sidekick will answer directly; other times one routes the request to another who summarizes the ticket or coaches the analyst. To the analyst, it feels effortless — one assistant, many skills. Under the hood, it’s real agent orchestration at work.

And because Sidekicks learn in the real world, analysts can shape them with quick feedback — a thumbs up, a thumbs down — making the system sharper over time.

The result: an always-on copilot that handles the messy middle of IT with speed, context, and clarity.

What’s cool about Ask Sidekicks

Three reasons this is harder than it looks

You’ve seen the flashy clips — point a chatbot at a PDF, sprinkle in five lines of Python, and suddenly it’s “the future of work.” 

But it’s harder than it looks. If you’re interested in what’s going on under the hood, here are a few of the hurdles that come at you fast when you try to make a feature like Ask Sidekicks real:

1. Real-time data migration

Analysts don’t want answers from “a few minutes ago.” They need the latest ticket state, chat message, or knowledge base update. Early versions of Sidekicks leaned on batch data from ClickHouse — good for reporting, terrible for live problem-solving.

We had to rip and rebuild: streaming APIs, rethought schemas, tighter payloads, faster refresh cycles. Every choice was a tradeoff between fidelity and responsiveness.

2. Handling massive context windows

Sidekicks regularly juggle 50,000+ tokens of context — the equivalent of a hundred-page document. Most models choke on that. Even when they don’t, latency and cost still spike. That’s a problem when analysts expect answers in under 10 seconds. Anything over 30 seconds? Game over.

So we had to apply some thoughtful strategies: chunking, compression, prioritizing context. Get it wrong, and Sidekicks surface stale or irrelevant details – and trust erodes instantly.

3. Orchestrating multiple agents

At first, analysts had to route questions to different specialized agents. The problem? They didn’t always know who to send it to — or how to get what they needed. Most of the time, they just had a question and wanted an answer.

So we redesigned Sidekicks to act more like an orchestrator — a front door into an ecosystem of assistants — so analysts no longer had to worry about who did what behind the scenes.

That introduced its own set of headaches: latency stacking when agents hand off, UX questions about how much to show, and coordination with other orchestrators. 

To tackle these headaches – we’ve built a fast classifier that can decide in ~300ms whether a query belongs to Sidekicks for general Q&A or should skip straight to another, more specialized agent. This keeps chat responses snappy and accurate.

On the orchestration side, we’re exploring a manager–worker model, where one agent maintains state and context while delegating work to others, which could give Sidekicks more control without bogging down the experience.

Ask Sidekicks in action

Building a feature is one thing. Seeing the impact is another.

Ask Sidekicks is used consistently, often in ways we couldn’t have predicted. And that’s the point: give analysts tools, and they’ll use them creatively to solve problems faster.

5 types of question our analysts asks: