Skip to content
60x
  • Company Brain
  • Knowledge Graph
  • Industries
  • About
  • Case Studies
  • Blog
Book a demo →Book a demo →
All posts
Team Alignment•13 Nov 2026•12 min read•Fergus McKenzie-Wilson

Faster alone, slower together

The alignment problem AI made worse, and how a company brain closes the information gap.

Key takeaways

  • Faster individual AI work can increase duplicated research and conflicting decisions across a team.
  • Shared memory helps people and agents start from current work, with dates, sources and permissions attached.
  • Measure repeated work and the time spent establishing facts; judgement and leadership still belong to people.

On this page

  1. Everyone is faster. Who is keeping the team aligned?
  2. Alignment was already expensive
  3. Then everyone got faster
  4. AI adds more places for context to drift
  5. What misalignment looks like up close
  6. Missing information and different judgement need different fixes
  7. Give the work a shared memory
  8. What a brain cannot settle
  9. Start where duplicated work costs most
  10. Sources
  11. Frequently asked questions
  12. Related reading

Everyone is faster. Who is keeping the team aligned?

I ask founders a simple question when I first meet them: what are your tasks today?

Two years ago, the answer was a list of tickets. Now, at companies somewhere past Series A, I often hear: "I've got this Claude thread, and this Claude thread, and this Claude thread." Some teams have drifted away from Linear and email. Everyone has locked in on their own agents. Output is up. A company brain has to keep the rest of the team in view.

Then I ask how they know two people aren't doing the same thing. Nobody has an answer. We see the cost in sales-to-delivery handoffs at consultancies.

That is the alignment problem I see in 2026. Every person is moving faster, while the company struggles to keep up with itself.

Three conceptual stages: silos between teams, between people, and between each person's AI threads. Purple threads represent duplicated work.
Conceptual illustration. AI can create another layer of fragmentation inside existing silos.

Alignment was already expensive

Coordination cost a lot before anyone had an AI assistant. In 2012, McKinsey estimated that interaction workers spent almost 20% of the week finding internal information or tracking down colleagues. Asana's 2023 survey of 9,615 knowledge workers put "work about work" at 58% of their time.

Atlassian's State of Teams 2025, based on 12,000 knowledge workers and 200 Fortune 1000 executives, found that one in two workers said teams had worked on the same things without knowing it. Only 7% of executives felt confident they knew how every team's work supported company goals.

Microsoft adds the texture. Its 2025 Work Trend Index found about half of employees and leaders described work as chaotic and fragmented. Among the top 20% of users by incoming ping volume, the average gap between interruptions during core working hours was two minutes.

These studies use different definitions and populations. They describe a familiar cost: when knowledge lives in separate heads, inboxes and threads, finding out becomes a job of its own.

Independent research findings on information search, work coordination, duplicated work, goal alignment and interruptions. These figures describe different populations and are not additive.
Sources: McKinsey (2012), Asana (2023), Atlassian (2025) and Microsoft (2025). Separate measures, not parts of one total.

Then everyone got faster

The individual gains from AI are real. The organisation doesn't collect those gains by default.

Faros AI analysed telemetry from more than 10,000 developers across 1,255 teams. In higher-adoption periods, developers completed 21% more tasks and merged 98% more pull requests. Review time per pull request rose 91%. At company level, Faros reported no significant correlation between AI adoption and improvement.

Google's DORA 2025 research, drawing on almost 5,000 technology professionals, found a positive relationship between AI adoption and delivery throughput, alongside a negative relationship with delivery stability. Its argument is that AI amplifies the systems around it.

These are observational findings about software teams. They echo what I hear in funds, consultancies and scale-ups: people can get far enough alone that they skip the alignment conversation.

One person described the change well. Previously, everyone would bullet-point a solution together before starting. Now you build it, and discover later that a colleague built something next to it.

Faros AI: tasks completed rose 21%, pull requests merged rose 98%, and review time per pull request rose 91%. No significant company-level correlation was reported; that does not mean a measured zero effect.
Faros AI, 2025. Comparisons between higher- and lower-adoption periods. Longer review time is the bottleneck, highlighted in purple.

AI adds more places for context to drift

Silos now sit inside silos

Each person has several threads, and every thread carries its own context. You can end up with five single sources of truth: two in one person's threads, one elsewhere, two more over here. Each makes sense on its own terms. None is guaranteed to reflect the others.

More participants create more possible connections

Brooks's communication-path model gives a group of n participants n(n − 1) / 2 possible pairwise connections. Ten people have 45. Add ten agents and the fully connected model has 190. Real teams do not maintain every connection, but the arithmetic shows how fast coordination can expand.

Microsoft now proposes a human-agent ratio as a management metric. In its 2025 survey, 46% of leaders said their organisations were already using agents to fully automate workflows or processes.

Agents have started asking each other

Nicolas Bustamante described his agent messaging a colleague on Teams, and the colleague's agent answering from its own context. The exchange makes the coordination question more urgent: where is the shared record, and what may each agent disclose?

A fully connected network has 45 possible links for 10 people, and 190 for 10 people plus 10 agents. This is an illustrative model, not measured communication traffic.
Communication-path model: Frederick P. Brooks Jr., The Mythical Man-Month (1975). The agent example is illustrative.

What misalignment looks like up close

These examples come from client and prospect calls over the past year. The organisations are anonymised.

Two documents, one methodology

A private equity fund and our team were defining a scoring methodology. A spreadsheet and a written document were both meant to be authoritative. Someone noticed they disagreed. The answer was "this one is more up to date": an uncomfortable sentence about a methodology that ranks investments.

The deep dive nobody knew about

The same fund's origination team flags a company in the CRM when research starts, so colleagues do not duplicate it. The flag exists because the duplication had happened before. Manual tagging was the defence.

Everyone grinding the same data room

An investor at a venture fund told me she had no idea what colleagues were researching. Everyone ran their own process. When a hot deal arrived, the team analysed the same data room in parallel and the firm paid for the same answer several times.

Three versions of the scope

In delivery projects, the statement of work can say one thing, the sales conversation another, and the engineers build to a third. An alignment call follows when someone notices. A shared record can expose the disagreement earlier.

The company that went back to tickets

A fast-growing AI startup we are onboarding told us alignment had become a theme at its current size. Its response was more discipline with the ticket tracker: a return to a practice the team had been leaving behind.

The org chart and the working map

When we map collaboration from permitted meetings, chat and email, the picture often differs from the org chart. People expected to work together may seldom talk. The gap is hard to discuss until somebody draws it.

Two threads, one shared record

What's the latest on the Acme deal?

Analyst A

Phase one: full scope.

Earlier scope document

Analyst B

Phase one: reduced scope.

Approved 4 Oct · Deal desk note
Each thread gathers its own context

Email, chat, documents and CRM queried one at a time.

Same question. Different versions.

Illustrative example. The analysts have the same source permissions. Source: 60x. Reduced motion is enabled; use the scene buttons.

Missing information and different judgement need different fixes

Sometimes two people disagree because they have seen different things. One read the update. One heard the client call. The disagreement can disappear once both see the evidence.

Sometimes they have seen the same evidence and weigh it in different ways. That is healthy judgement, and it is where a team earns its pay.

In my experience, the information gap is more common, and it crowds out the useful disagreement. Meetings go on establishing facts before anyone can debate a decision.

The aim is a record that keeps itself current, so the meeting can debate what the evidence means instead of spending the hour establishing it.

A company brain helps close the information gap. The judgement stays with the team.

Give the work a shared memory

A company brain holds what the team knows from the places work happens: chat, meetings, email, documents, tickets and the CRM. People and their agents can ask it anything they are allowed to see, from one place.

Start each thread from current company context

Every agent starts from the same company record instead of each rebuilding it from scratch. Where two people have the same access, their answers draw on the same evidence. Where access differs, the answers differ too.

Write completed work back

When an analyst finishes research, the findings go back into the brain with their supporting evidence and access rules. The next authorised person picks up the analysis instead of repeating it. One fund we work with applies its investment thesis across its universe this way, updating it as people learn something proprietary. The alignment improvement mattered more than the speed improvement, which I did not predict.

Keep dates and sources attached

Your team can see when a fact changed in its source, and where it came from. Conflicting versions show up side by side. A date alone does not decide which document is authoritative, but it gives the team evidence to resolve the disagreement.

Give the same thing one identity

The same company, person or project shows up as one record, even when each system names it differently. People discussing "the Acme deal" work from the same picture.

Tell people before they collide

A daily or weekly brief can put the deal desk's latest position in front of the account lead before the client call. The brain tells people about relevant changes as well as answering questions.

Make the collaboration map visible

Within the information a reader is allowed to inspect, a collaboration map can reveal working relationships the org chart misses. It gives leaders a starting point for a conversation about handovers.

Keep private context private

Alignment does not require a company-wide view of everyone's inbox. Each person sees only what the source systems already let them see. Private personal context stays separate from shared company knowledge. Read how the brain knows who is asking.

Shared work sources feed one company brain. Authorised people, agents and daily briefs read through that layer, and findings can be written back with evidence and permissions.
Source: 60x. Shared context remains scoped to the reader's access.
The same twelve illustrative people appear in an org chart and a collaboration network. A cross-team working pair and a missing expected connection are highlighted.
Illustrative network, not client data. Reporting lines and working relationships describe different things.

What a brain cannot settle

It records what was written and said in the systems it reads. Knowledge that never left someone's head stays there. Transcripts and chat capture some of it, but never all of it.

Capture also changes behaviour. Someone joining a company where calls were recorded and shared described the consequence: sensitive conversations moved somewhere that was not tracked. Permissions reduce that pressure because private conversations can stay private. The tension still deserves attention when designing a rollout.

And if two partners read the same evidence and disagree, the brain can help them reach the evidence faster. It cannot make their decision for them.

Start where duplicated work costs most

For a fund, that may be deal research. For a software company, it may be the handover between sales, product and engineering.

Connect the channels where decisions happen. One recent pilot began with team chat, the meeting assistant, the wiki and the issue tracker. Those four covered much of the decision trail.

Start a daily or weekly brief early. Then measure how often people begin the same work, how long it takes to establish which version is current, and how many meetings exist to assemble facts.

If the brain is useful, those costs should fall. The meetings worth keeping are the ones where people make decisions from shared evidence.

With 60x, your first company brain use case is live in two weeks, with scope and access agreed before we start. Choose 60x Cloud with EU or US residency, or Your Cloud inside your own Azure, AWS or GCP cloud environment. See the installation plan. Book a demo.

Sources

  1. McKinsey Global Institute, The social economy, July 2012.
  2. Asana, Anatomy of Work Global Index 2023, March 2023.
  3. Atlassian, The State of Teams 2025.
  4. Microsoft, 2025 Work Trend Index: The year the Frontier Firm is born, April 2025. See its methodology for the interruption measure.
  5. Faros AI, The AI Productivity Paradox Report 2025. Analysis reflects data as of June 2025.
  6. Google Cloud, Announcing the 2025 DORA Report, September 2025.
  7. Frederick P. Brooks Jr., The Mythical Man-Month, 1975. Pairwise-path calculations are illustrative.
  8. Nicolas Bustamante, post on X, September 2026.

Client and prospect examples are anonymised. They describe our experience, rather than controlled research.

Frequently asked questions

Why can AI make a team less aligned?

People can produce more in separate agent threads while losing sight of work and decisions elsewhere. The result can be duplicated research, conflicting documents and extra time spent checking which version is current.

How does a company brain give AI agents shared memory?

It connects the records where work happens and makes completed findings available with their evidence, dates and access rules. The next person or agent can start from that work instead of rebuilding the same context.

How do you measure team alignment after introducing AI?

Track how often people begin the same work, how long it takes to find the current version and how many meetings mainly reconstruct facts. Use the shared evidence to support decisions while keeping responsibility with the team.

Related reading

  • Sold, scoped, shipped

    Why forward-deployed teams drift out of sync with sales and engineering, and how a company brain keeps them aligned.

  • The $10,000 analyst

    Why AI token bills explode at research firms, and how a company brain brings them down.

  • What you own when you own your AI

    Open weights make the model yours. Your company's context, and the tools your people use it through, make the AI yours.

Author

Fergus McKenzie-Wilson

Fergus McKenzie-Wilson

Founder

Pioneered Multi-Agent AI in 2023, transforming a consultancy into AI-native SaaS. Serial entrepreneur and exited founder; co-founded Hydra Drones at university and exited after 4 years. Awarded an Innovate UK grant and received multiple innovation awards from the Austrian government. Fergus architects and leads the engineering of 60x's agentic infrastructure and proprietary Knowledge Graph design.

Want help turning a research thesis into a working system?

We build the software, workflows, and knowledge layers that make AI useful inside real operating teams.

Book a demo →Book a demo →
60x

We help enterprise leaders deliver AI outcomes.

Company BrainKnowledge GraphWhat We KnewPermissionsTrust
Private EquityFinancial ServicesConsultancies
AboutCase StudiesJobs
BlogContactTermsPrivacyCompanies House

Solutions

Company BrainKnowledge GraphWhat We KnewPermissionsTrust

Industries

Private EquityFinancial ServicesConsultancies

Company

AboutCase StudiesJobs

Resources

BlogContactTermsPrivacyCompanies House
Cyber Essentials Certified
Unicorn Mafia

© 2026 60x.ai. All rights reserved.

PrivacyTerms of Use