Daniel Bullock

Founder & CEO of Neotask

Human direction.
Agent execution.

I’ve spent years building autonomous software. With Galaxsio, users purchased more than $40 million in product through systems I built.

Advances in AI let me apply that experience to a much wider range of work. I’m now building Neotask and the tools that help agents plan, use apps, coordinate, and execute within defined limits.

My coding process Projects About me

What I’m building and exploring

Neotask

My current company brings agents, connected apps, and ongoing work into one environment. People set the objectives and boundaries; agents carry out the work.

About Neotask

Intervieux

I am building Intervieux with my brother in partnership with Level Up Ventures. AI interviews, career support, and hiring tools connect candidates with employers.

About Intervieux

Agent development

The software factory I use across my projects: orchestration, reusable procedures, memory, and verification. Neomax coordinates the coding agents behind it.

Inside my process

Ideas & experience

What I’m learning about AI, autonomy, and building software businesses. A place to get to know the founder and follow the thinking behind the work.

Visit the blog

AI, growth, and the work people do.

I think AI can let businesses grow in ways that were previously out of reach. A team that can serve more customers or develop a new service has more room to invest in people and take on work it could not afford before.

I want that growth to keep people employed, create opportunities, and expand the scope of what a business can do. The decisions businesses make about that new capacity will shape the jobs that follow.

Helping people find those opportunities is part of my work too. With Intervieux, we are using AI to connect the effort someone puts into getting hired with the information an employer needs to make a decision.

Building Intervieux

In partnership with Level Up Ventures

I co-founded Intervieux with my brother, and we are actively developing both sides of the hiring process.

For people looking for work

Job discovery, resumes, AI interviews, simulations, and technical challenges help candidates prepare, demonstrate their ability, and get connected with employers.

For teams making a hire

An AI-powered applicant tracking system brings interviews, resumes, and assessments into the hiring pipeline. Employers can review the evidence, coordinate next steps, and move candidates toward an offer.

More about Intervieux Visit Intervieux

The system around the agents

Planning, execution, verification, and recovery, all operating within defined boundaries.

Workflow architecture

Human direction

Objectives · priorities · constraints · success criteria

Shared context

Code · documentation · memory · skills · SOPs

Orchestration

Coordinate the work and decide what can run next.

Provider & model routingScheduling & queuesTask & session stateCapacity & budgets

Results, failures, and new context feed the next cycle

Enforced across the workflow

Rules & policies
Permissions
Execution guards
Approval gates
Spend limits
Timeouts & stop conditions

Autonomy

Continue inside the approved scope. Escalate exceptions with the evidence needed for a decision.

Explore a stage

Task planning

Turn the objective into tasks with explicit dependencies, owners, and acceptance conditions. The task graph determines what can run in parallel and what must wait.

Building agent systems

What useful autonomy depends on

An agent needs more than a task description to work on its own. These are the capabilities I work on across Neotask and the development systems behind it.

Context & memory

Keep the brief, decisions, and current task state available across sessions. An agent resuming a job needs to know what changed, what passed, and what remains.

Agent workspaces, session history, durable task records

Tools & integrations

Give agents specific ways to act: read a document, update a task, inspect a build, or work in an app. Define access at the tool and account level so the allowed scope is enforceable.

MCP providers, app connections, authenticated tool calls

Coordination

Split work along boundaries that can be checked independently. Track dependencies and ownership so concurrent agents can make progress without overwriting each other’s changes.

Worker dispatch, task dependencies, isolated worktrees

Rules & operating limits

Define what can run autonomously and what needs approval. Put spending, execution scope, and stop conditions into the system that runs the agent.

Approval policies, budget controls, worker scopes

Recurring execution

Turn a recurring job into a procedure with inputs, a schedule, and an expected result. Keep its execution history so a missed run or repeated failure is visible.

SOPs, plans, schedules, run history

Verification & recovery

Check the result where it will be used. Preserve enough state to retry or resume a failed step, and keep deployment and rollback instructions alongside the work.

Tests, browser checks, release evidence, recovery records

Inside the development process

Behind the product

I work with 10+ Claude Max and Codex subscriptions, alongside Kimi Code. My software factory connects the agents, tools, and repeatable workflows I use across my projects.

The engineering challenge is keeping all of that work coherent. A change to an integration can affect authentication, a gateway tool, the app interface, and documentation. Those pieces need a shared definition of done.

10+Claude Max & Codex subscriptions

+ Kimi CodePart of the development setup

01 / Scope

Make the outcome testable

Write down the affected repositories, expected behavior, constraints, and proof required. Break the work into parts with clear ownership and dependencies.

02 / Execute

Coordinate the workers

Route tasks to an appropriate provider and model, isolate concurrent changes, and preserve task state when a worker stops or an account reaches a limit.

03 / Verify

Check the complete path

Test the code and the user flow. For an app integration, that includes connection, authorization, tool execution, and how the result appears in the interface.

04 / Repeat

Keep what worked

Save the procedure, checks, and recovery steps. The next provider integration or upstream release starts with a workflow that already records what needs checking.

Development infrastructure

Neomax

My Rust orchestrator coordinates coding agents, accounts, models, sessions, and worktrees. Managed workers keep durable task state, with scoped routing and recovery when execution is interrupted. It supports Claude, Codex, OpenCode, Kimi, and Grok adapters.

More about the workflow →

Practical starting points

Designing work an agent can finish

A few useful questions to answer before adding autonomy to a workflow.

What does “done” look like?

Describe a result that can be inspected. “Research this account” leaves a lot open. “Produce a brief with source links, unanswered questions, and the next action” gives the agent a deliverable and gives you a way to judge it. Include where the result belongs and what to do when information is missing.

Which tools and permissions does the task actually need?

List the systems the agent must read and the changes it may make. Reading a CRM record, drafting an update, and sending that update are separate capabilities. Give the agent the access needed for its job, and make the approval boundary explicit for actions outside that scope.

When does splitting work across agents help?

Look for tasks with independent inputs and outputs: investigate separate components, implement changes in separate areas, or review a completed change. Work that repeatedly edits the same files needs sequencing or explicit ownership. More concurrent workers also create more coordination work.

How does the workflow handle interruption?

Persist the current task, completed steps, and any external actions already taken. Before retrying, check whether the previous attempt succeeded. A process should be able to resume without sending the same message twice or repeating a change that already landed.

What should trigger a person’s attention?

Define the exception cases up front: missing access, contradictory instructions, exhausted budget, repeated failures, or a decision outside the approved scope. Include the evidence a person needs to resolve the issue so the agent can continue from a clear decision.

Projects

What I built, the problems involved, and my role in the work.

All projects

Neotask

I’m building Neotask to connect the parts of agent work that usually live in separate places: the objective, the conversation, the tools, the tasks, and the result.

Founder & CEO

Intervieux

I co-founded Intervieux with my brother, and we are actively building it in partnership with Level Up Ventures. We are connecting the work of finding a job, demonstrating skills, and making a hiring decision in one platform.

Co-founder, with my brother

BetterSlip

I built BetterSlip from scratch to make sports betting markets easier to compare. The platform scanned millions of odds each day across more than 140 sportsbooks, with data arriving at one-to-three-second latency.

Sole developer & team lead

Galaxsio

I built Galaxsio from scratch as a consumer-goods purchasing automation platform. It grew into two full software versions, more than 20 retailer integrations, and a product used by over 2,000 people.

Co-founder & CTO

Watchtower Protection

I built Watchtower Protection for Doris Talent Agency, working closely with legal counsel and agency management. The product needed to fit creators’ needs while giving the agency an evidence-based enforcement workflow.

Sole developer

Rockford Financial

Rockford Financial was my self-funded reselling business. I was the sole owner and operator, building the automation that supported purchasing and running the business around it.

Founder & CEO

KDG Slack

I co-founded KDG as a paid reselling intelligence community. I built monitors tracking thousands of sites, including the Shopify ecosystem, plus Slack alerts and Chrome extensions for subscribers.

Co-founder

KDG Slack convention booth — KDG at a sneaker convention

DNA IO

After my freshman year at the University of Iowa, I took time away from school and taught myself to code. DNA IO, Nike SNKRS purchasing automation software built with one partner, became my first licensed product.

Co-builder, with one partner