Your AI Is Only as Smart as Your Company Memory

A surprising amount of what a company knows never makes it into a document. It lives in Slack threads, meeting discussions, email exchanges, and quick conversations between people who understand why a decision was made. That has always created a knowledge-management problem. With AI agents, it becomes an infrastructure problem too.

This week, Slack CPO Jaime DeLanghe described how Slack is adapting workplace communication for teams where humans and agents work together. One of Slack's key lessons is that agents become more useful when employees work openly because conversations contain context that formal documentation often misses. Slack recommends putting more work in public channels, connecting agents to sources such as meetings, email, calendars, and documents, and creating explicit handoffs between people and agents.

The implication for technology companies goes well beyond Slack. Two companies could deploy the same model and give employees access to similar AI tools, yet get very different results because one has made its institutional knowledge much easier for AI to understand.

Companies have spent years thinking about internal communication primarily as a collaboration problem. Information needed to reach the right people quickly enough for them to make decisions and keep work moving. Agents give those conversations another purpose because they can turn everyday communication into reusable organizational context.

Consider a technical decision made during a product discussion. The final choice might eventually appear in Jira or a project document, but the reasoning behind it may remain buried in a Slack thread. Engineers discussed three approaches, identified an infrastructure constraint, rejected two options, and settled on the third. For a colleague who participated in that conversation, the context is obvious. An agent that can only see the final ticket is missing most of the useful knowledge.

Giving AI access to the reasoning behind decisions can make the same underlying model much more useful.

The AI industry has spent enormous energy talking about proprietary data. Companies with unique customer data, operational information, or domain-specific datasets are expected to have an advantage when building AI products. Internal context deserves similar attention.

Every organization accumulates years of decisions, exceptions, tradeoffs, customer insights, technical discussions, and operating knowledge. Much of it is scattered across communication tools and people's memories instead of organized into a clean knowledge base.

Agents create an incentive to make more of that context accessible. Conversations, documents, calendars, meetings, and systems of record can collectively become a memory layer that helps AI understand how the company actually operates. Information architecture is becoming part of AI architecture.

Slack's recommendation to move more work into public channels is especially interesting because it connects organizational behavior directly to AI performance.

A decision made in a private message may help two employees today. A decision discussed somewhere accessible can inform employees and agents months later, preserving the reasoning as reusable institutional knowledge.

Permissions still matter, particularly for sensitive employee, customer, financial, and strategic information. Companies need clear boundaries around what agents can retrieve and who can access the results. Within those boundaries, organizations that make useful context easier to discover give their agents more material to work with.

AI adoption is exposing an old organizational weakness: companies often know much more than their systems can retrieve.

Better context also changes how companies can structure human-agent workflows. Slack's model points toward explicit handoffs where agents prepare or execute parts of the work and return important decisions to people.

An agent might gather the history of a project before a meeting, prepare relevant documents, or identify unresolved decisions. Afterward, it could capture outcomes and update the systems where future employees and agents will find them. Humans remain responsible for decisions that require judgment, while agents help preserve and carry context between stages of the work.

Over time, each workflow can leave behind better organizational memory for the next one. That could also reduce a familiar source of wasted time as employees repeatedly search for decisions, reconstruct why something happened, ask colleagues for background, or redo research because the original context is difficult to find.

This issue of Redeployed is brought to you by Tecla Labs: As AI inference gets cheaper, companies can rethink how their AI products are built. More affordable models make it practical to route work across different systems, run agents in parallel, add verification, and use more intelligence throughout a workflow. Building these architectures requires engineers who understand AI infrastructure, model orchestration, and production software. Tecla Labs designs and builds AI systems for companies that need this expertise now, so teams can ship flexible AI products without locking themselves into a single deployment model.

Making company knowledge machine-readable creates a much larger surface for permissions and privacy. An agent that can search across Slack, meetings, documents, email, and calendars may have access to far more organizational context than any individual employee should see.

Information quality matters too. Companies change direction, and old decisions are superseded. An agent that retrieves a confident six-month-old discussion without understanding that the policy changed last week can turn institutional memory into institutional confusion.

There is also a human consequence. Employees may communicate differently when they know conversations can become inputs for AI systems. Organizations will need clear expectations about what gets captured, how it is used, and where private communication remains private. Poor information hygiene will not disappear when agents arrive. AI can amplify it.

AI readiness increasingly includes the quality of a company's internal knowledge environment. Leaders should look at where important decisions happen, whether the reasoning behind them survives, how easily information can be retrieved, and which systems contain context employees repeatedly need to reconstruct.

Buying a more capable model may produce limited gains when the agent cannot access the information required to do useful work. Improving the context layer can make existing AI systems substantially more effective.

This work cuts across engineering, IT, security, and operations. Technical teams need to solve integrations, identity, permissions, retrieval, and monitoring. The organization also needs clear practices for where information belongs and how important reasoning gets preserved.

As workplace agents spread, their performance will depend heavily on the environment around them. Organizations with fragmented conversations, inaccessible decisions, and weak documentation will give agents an incomplete picture of how the business works.

Companies that preserve reasoning and make useful context accessible can get more from the models they already have. Over time, that advantage can compound as each decision, project, and workflow contributes more useful context for future work.

Your company's communication habits are quietly becoming part of its AI stack.

Connect With Other Technology Leaders

If you want to connect with other technology leaders having real conversations about AI and how it is changing business, check out GILD Curated Circuit.

More to come…

✔️ Goldman’s engineers have a new challenge: turning AI agents into firm insiders — Business Insider

✔️ OpenAI is building AI agents for everything. Will everyone use them? — TechCrunch

✔️ Forget chatbot training. AI’s next big data grab is about learning how humans work. — Business Insider