What we're building toward: the Avery NXR vision through 2028

We've written 230 pieces of content (this is the 230th). Most are about the present — what Avery NXR does now, who it's for, how it compares.

This piece is about the future. What we're building toward over the next 2-3 years. The bet we're making about how AI agents evolve and how Avery NXR fits.

We'll be transparent about what's roadmap and what's vision. Roadmap is what we're committing to ship. Vision is what we're betting on but haven't fully committed to.

The 2028 thesis

Our bet: by end of 2028, the AI agent landscape will look like this.

Operational AI becomes infrastructure. Like email, like CRM, like analytics. Every company with more than ~20 employees has multiple agents running. The question isn't "should we have agents" — it's "which agents, configured how."

Local-first becomes the default architecture for operational AI. The cost economics + privacy economics + reliability economics all push toward local. Cloud-LLM platforms continue to exist but for different use cases (frontier reasoning, novel one-offs, conversational AI).

Agent ops becomes a recognized profession. Like DevOps in the 2010s. Established titles, career paths, certifications, conferences, communities.

Audit + governance + safety become table stakes. Regulatory pressure makes these required, not differentiators. Platforms without strong governance support don't survive.

Multi-agent compose becomes the marquee feature. Single-agent platforms are commoditized. The valuable thing is orchestrating multiple agents.

The market consolidates. Currently fragmented (dozens of platforms across categories). By 2028, we'd expect 4-6 dominant platforms across the various categories (conversational, operational, autonomous, augmentative).

Avery NXR's bet is to be one of the dominant platforms in the OPERATIONAL category by then.

What we're building (roadmap)

Specific features committed to ship in the next 12-18 months:

Better multi-agent orchestration.

Sub-agents exist today. We're investing in making complex multi-agent workflows easier to build, debug, and maintain. Visual representation of multi-agent flows. Better state management across agents. Faster execution of complex graphs.

Connector expansion.

Currently 63. Targeting 100+ by end of 2027. Focus on: → Salesforce (enterprise gap) → Microsoft Teams (Slack alternative) → Google Sheets (bidirectional sync similar to Excel) → Airtable → Trello, ClickUp → Industry-specific connectors based on customer requests

Better local model support.

Standard support for newer models as they release. Hardware-aware recommendations. Specialized models for specific tasks (extraction, classification, summarization).

Enhanced audit + governance features.

Compliance regime-specific audit outputs (SOC 2, GDPR, HIPAA, etc.). Better visualization of agent decision patterns. Bias monitoring dashboards. Configurable retention policies.

Onboarding improvements.

Better first-week experience based on first-100-customer learnings. Improved templates. Guided agent building for non-engineers.

Enterprise deployment features.

SSO improvements. Better multi-tenant deployment patterns. Role-based access enhancements. Audit log integration with SIEMs.

Better debugging tools.

Built on the foundation in [post 227]. Better visualization of multi-step workflows. Automated regression detection. Suggested fixes based on common patterns.

What we're investigating (vision)

Areas we think matter but haven't fully committed to:

Specialized models for operational tasks.

The local model layer (Qwen, DeepSeek, Llama, etc.) is general-purpose. There may be value in specialized models tuned for specific operational tasks — classification, extraction, drafting in specific styles.

We're investigating whether to develop or curate specialized models. Decision not made.

Marketplace for agent templates.

Currently we have 7 pre-loaded templates that we maintain. Some platforms have marketplaces where community contributes templates.

We're skeptical of marketplaces (covered in [post 166]) but watching as customer base grows. May change view.

Agent observability platform integration.

Some teams want their agent operations data flowing into existing observability platforms (Datadog, New Relic, etc.). We're investigating whether to build integration vs. ask customers to do it themselves.

Voice integration (limited scope).

We've been clear that we're not building voice agents. But we're investigating limited voice INPUT for agent configuration — talking to set up an agent rather than typing.

This isn't competing with Sierra/Decagon. It's specifically about reducing setup friction for non-typing-friendly users.

Mobile experience.

Currently Avery NXR is desktop-focused. We're investigating what (if anything) a mobile companion experience should look like. Probably not a full mobile builder. Possibly a notification + audit ledger viewing experience.

Plugin ecosystem.

We get requests for third-party plugins (custom step types, custom UI components, custom integrations). We're investigating whether to build a plugin system or stay closed.

What we're NOT building

Reaffirming from [post 166] with some updates:

→ Not a chatbot. Still no. → Not a cloud LLM of our own. Still no. → Not coding agents. Still no. → Not voice agents. Still no (limited voice input is different). → Not multi-tenant cloud SaaS as primary product. Still no. → Not a marketplace yet. Re-investigating but not committing. → Not consumer-facing AI. Still no. → Not fine-tuning workflows. Still no.

These scope decisions hold. They define what we can do well.

Pricing evolution

We expect pricing to stay roughly stable in shape. Specifics may evolve:

→ Free Desktop: remains $0 forever → Pro: currently $29/user/month. May add tiers as features expand (Premium Pro with extras vs. base Pro). → Enterprise: custom, scaling with feature set + user count + support level.

We don't see fundamental pricing model changes. Flat per-user is what we believe in.

Geographic expansion

Currently most customers are US-based. We're investing in expansion:

→ EU support. GDPR compliance documentation. EU-region cloud deployment options. Localization for German, French, Spanish (in priority order). → APAC support. Japan + Australia first. Local language support over time. → Other regions. As demand emerges. Latin America, Middle East, Africa over longer horizon.

Local-first architecture helps with international expansion — data residency requirements are met by design in many regions.

What we're investing in for our own team

Some org-level decisions:

→ Hiring more engineers, particularly with strengths in distributed systems, security, observability → Building customer success function (currently founder-led; needs to scale) → Building developer relations / community function (for the agent ops practitioner community) → Strengthening compliance + legal capability as we move into regulated industries

These investments shape what we can ship.

What could change our roadmap

We're flexible. Things that could shift priorities:

Major model capability changes. If frontier models or local models change dramatically, our priorities adjust.

Regulatory changes. New AI regulations could push compliance features forward.

Customer demand patterns. What customers actually want vs. what we predict may differ.

Competitive movements. If competitors release things that change customer expectations, we may respond.

Macroeconomic factors. Recession would shift priorities toward retention + cost efficiency.

The roadmap above is our best plan. It will evolve.

What customers can rely on

Even with evolution, some things we commit to:

→ Free Desktop tier stays free → Local-first architecture remains the default → Audit ledger stays as foundational, not feature add-on → Per-user flat pricing for Pro tier → Transparent development (we'll keep writing about what we ship) → Customer data stays with customers (we don't take data into our cloud for desktop tier)

These are architectural commitments. They define who we are.

What this means for buyers today

If you're considering Avery NXR right now:

You're buying for current capability + future trajectory. The current capability is strong (170+ paying customers, 220 pieces of content explaining what we do). The trajectory is local-first operational AI continuing to mature.

The bet you're making. That local-first AI agents become important infrastructure. That we'll be one of the dominant platforms in that category. That the architectural choices we've made age well.

What you'd be unhappy about. If you need conversational AI, autonomous task agents, browser automation, fine-tuning workflows, or consumer features — we're not for you.

What you'd be happy about. If you need operational AI for recurring workflows with cost predictability, data residency, audit transparency, and team-level deployment — we're well-positioned.

A note on this being the 230th piece

We've written a lot. This isn't slowing down.

Why we keep writing:

→ It's how we explain who we are to people considering us → It's how we think clearly about our own positioning → It's how we build trust over time vs. one-shot marketing → It's how we earn the right to be considered alongside larger competitors

If you've read meaningful portions of these 230 pieces, you know what we're building. The content is the longest-form sales pitch we have. It's also the most honest.

We'll keep writing.

The principle

Building a platform takes years. The architectural choices made in early years compound for decades.

Avery NXR has made its architectural choices: local-first, operational-focused, flat-priced, audit-transparent, non-engineer-accessible.

Over the next 2-3 years, we'll continue building on these foundations. The features expand. The architecture remains.

If the architectural choices are right (we believe they are), the platform compounds in value. If they're wrong, we'll learn and adjust.

That's the bet. 230 pieces in. Many more to go.

→ avery.software — Free Desktop tier. The platform built with a 5-year view, not a 5-quarter view.